Paper Submission - NPS Australia Submission System

Do Macroeconomic Predictors Improve Bangladesh Inflation Forecasts? A Leakage-Safe Econometric and Machine-Learning Evaluation, 2010-2025 Abstract- This study evaluates whether lagged macroeconomic indicators improve

This study evaluates whether lagged macroeconomic indicators improve one-month-ahead forecasts of Bangladesh headline inflation beyond strong persistence-based benchmarks. A monthly data set covering January 2010 to December 2025 is assembled from public sources and contains year-on-year inflation, BDT/USD exchange-rate changes, the policy rate, Brent oil-price changes, and annual real GDP growth. To prevent information leakage, annual GDP is introduced only as the previous year’s realized value, and forecasts are evaluated through an expanding-window design on an untouched January 2023-December 2025 holdout. The analysis combines augmented Dickey-Fuller and KPSS stationarity tests, Ljung-Box diagnostics, a distributed-lag regression with heteroskedasticity-and-autocorrelation-consistent standard errors, Granger predictive-causality tests, SARIMAX, autoregression, Ridge regression, Random Forest, Extra Trees, Gradient Boosting, and XGBoost. Inflation is highly persistent, with a lag-one autocorrelation of 0.964. Exchange-rate lags are jointly significant in the explanatory model (Wald p < 0.001), whereas oil-price lags are not and policy-rate changes are only borderline significant. SARIMAX records the lowest RMSE (0.543 percentage points), but its improvement over the last-month benchmark is not statistically significant. Complex machine-learning models do not outperform the parsimonious benchmarks. The results indicate that exchange-rate information matters for inflation dynamics, but most short-horizon forecastability originates from inflation persistence. The paper demonstrates why chronological validation, strong naive baselines, and careful publication-timing controls are essential for credible macroeconomic machine learning.

Dynamic Performance Analysis of a 30kW PV System using P&O-Based Maximum Power Point Tracking

In this paper a detailed dynamic performance analysis of a 30-kW PV system using Perturb and Observe (P&O) maximum power point tracking (MPPT) algorithm is presented. As solar PV is increasingly integrated into today’s power networks, good MPPT is critical to harnessing maximum power. Although simple and inexpensive to implement, conventional P&O has a few significant disadvantages: oscillations around the maximum power point, loss of the direction of the power point tracking with fast changes in the solar irradiance, and compromising the direction of the tracking to speed up and increase the accuracy of the tracking. The proposed system is composed of PV array, DC-DC boost converter and P&O controller. The results of the simulations are compared and studied under the assumption of step changes in irradiance and load. The dynamic performance parameters like convergence time, oscillations under steady state condition, tracking efficiency and power quality are studied in detail. From the results presented, the performance of the standard P&O algorithm in steady state error tracking performance is good, while the performance in terms of dynamic response is limited, so there is a need to improve the performance of the dynamic response by using better variants of the P&O algorithm to achieve good performance in practice.

Detecting Financial Fraud Rings: Graph Motifs, Learned Detection, and a Base-Rate-Normalised Evaluation

1. Benchmark hygiene changes the answer. AMLworld’s final week has laundering at 59% purity against a 0.1% norm, holding 12.7% of all positives. A standard chronological split puts it in test. Removing it moves the graph-over-tabular advantage from 2.37× to 5.43×.

2. Precise graph queries degrade a GNN when injected as features. Detectors at up to 15.2× lift standalone cost 21% average precision and doubled seed variance as node attributes. Mechanism: they fire on 1.2% of accounts, so the columns are ~99% zero and dilute denser signal. This is a boundary condition on Blanuša et al., who report the opposite — their patterns attach to transactions and feed a tree ensemble, which can branch on a sparse indicator for free.

3. Two metadata columns beat 33 graph-derived ones. Ownership linkage — how many accounts share a legal owner — gave +10.7% AP and cut variance fourfold. Dense and semantic outperformed sparse and structural.

4. Base-rate-normalised lift, and the retention split. AP is uninterpretable across subsets of differing prevalence. Applied to leave-one-typology-out, recall retention is 93.5% but lift retention 45.5% — the model still finds an unseen coordination pattern, but ranks it far less confidently. Either figure alone misleads.

5. Empirical threshold selection for motif queries. Sweeping stored values rather than guessing more than doubled lift on three of four detectors — fan-out 6.2× → 15.2×, fan-in 2.7× → 12.9×

Academic Citation and Trend Forecaster: A Simple Graph Neural Network Approach for Venue Prediction and Collaboration Forecasting

A working end-to-end graph pipeline on a large real-world academic dataset — builds a heterogeneous graph (papers, authors, institutions) directly from OGBN-MAG’s raw files (736K+ papers, 5.4M+ citation edges), rather than a toy/small dataset.
A direct, controlled comparison of graph scale vs. architecture — full-graph GraphSAGE (mini-batch trained) vs. subgraph-sampled GAT, on the same task and dataset, showing that graph completeness (26.14% vs 9.24% accuracy) matters more than the choice of aggregation method (mean vs. attention). This is a useful, generalizable finding for anyone doing GNN work at scale.
A working link-prediction pipeline for collaboration forecasting — an encoder-decoder GraphSAGE model that scores author pairs, reaching 0.9998 ROC-AUC.
Methodological honesty as a contribution in itself — the paper explicitly flags that the near-perfect link-prediction AUC is inflated by random negative sampling and randomly-initialized author features, rather than presenting it uncritically. This kind of transparent limitation analysis is genuinely valuable and often missing from student/applied ML papers.

AI-Assisted Learning Tools in Improving Programming Skills Among University Students: A Case Study

This study investigates the impact of AI-assisted learning tools on programming skills among university students. It combines secondary analysis of an education dataset with a synthesis of recent computing-education research to examine changes in learning quality, learning time, stress, programming task performance, code comprehension, and debugging speed. The study also identifies the risk of over-reliance on AI tools and highlights the importance of structured pedagogical guidance for effective and responsible integration of AI in programming education.

Analysis and Prediction of Digital Distractions’ Impact on Academic Performance in Online Learning Using Machine Learning

This study combines survey-based behavioral analysis with supervised machine learning to predict the perceived academic impact of digital distractions and compares four classification models to identify the best-performing approach.

Integrating Deep CNN Models for Multilingual Sign Language Recognition: A SignLink-Based Approach for Bengali and English

Sign language is a vital medium of communication
for people who are deaf or hard of hearing. While much
research has addressed widely used systems such as American
Sign Language (ASL) and British Sign Language (BSL), Bengali
Sign Language (BdSL) remains far less explored. To help close
this gap, this work proposes a bilingual Bengali and English
sign language recognition system that applies deep learning for
stronger multilingual gesture interpretation. Two publicly avail
able datasets were merged. The first contains English letters from
A to Z, digits from 0 to 9, and a space symbol, while the second
holds 38 Bengali alphabet gestures. After the datasets were
combined and augmented to balance the class distribution, the
final collection held 75 classes and 112,493 images in total. Eight
deep learning models were evaluated, comprising six pre-trained
architectures and two custom networks. Among them, AlexNet
gave the best standalone test accuracy of 96.99%. To raise
accuracy further and reduce overfitting, a hybrid model named
SignLink was built by joining AlexNet with MobileNetV2 and
Xception, both of which generalize well. The hybrid reached a
test accuracy of 98.93%, surpassing every individual architecture.
The proposed system shows reliable bilingual recognition and
supports inclusive communication technology for linguistically
diverse and low resource settings.

VeriFake: A TF-IDF and Ensemble Learning Framework for Web-Based Fake News Detection

The rapid circulation of fabricated news has become a serious problem for society, since it spreads confusion,
reinforces bias, and erodes public confidence in trustworthy
information. Because digital platforms and social networks allow
misleading stories to travel far more quickly than before, they
often shape opinions, awareness, and everyday decisions. Reliable
and automated detection tools are therefore needed to limit
this damage. This study presents a supervised machine learning
framework that recognizes fake news directly from website text.
Term Frequency Inverse Document Frequency (TF-IDF) is used
for feature extraction, and thirteen classifiers are trained and
compared, namely Random Forest (RF), Decision Tree (DT),
Bernoulli Naive Bayes (BNB), Multinomial Naive Bayes (MNB),
Support Vector Machine (SVM), Linear Support Vector Classifier
(LSVC), K-Nearest Neighbor (KNN), Logistic Regression (LR),
XGBoost (XB), Multilayer Perceptron (MLP), Extra Trees (ET),
AdaBoost (AB), and Gradient Boosting (GB). The framework is
evaluated on a balanced Kaggle collection of 20,800 articles, split
evenly into 10,400 genuine and 10,400 false items. Random Forest
records the best accuracy at 99.23%, with LSVC and Decision
Tree following closely, while the remaining ensemble and boosting
methods also stay above 95%. For practical use, a secure Django
based web application was built so that users can submit news
text and receive an immediate prediction. The results confirm
that tree based ensembles and linear classifiers, when paired
with TF-IDF features, remain highly effective for monolingual
fake news detection, and the deployed platform shows how these
models can be turned into usable tools against misinformation.

Automated Code Smell Detection for Software Quality Assurance Using a Web-Based Machine Learning Framework

Code Smells are indicators of structural weaknesses
in software design and implementation that can reduce maintainability, scalability, and readability. To improve software quality
and support efficient maintenance, this study proposes a machine learning approach for the automated detection of Critical
Threshold Rule (CTR) violations, focusing on the Long Method
and Large Class smell types. In the feature selection stage,
six models, namely Decision Tree (DT), Random Forest (RF),
Logistic Regression (LR), Support Vector Machine (SVM), Multilayer Perceptron (MLP), and Stochastic Gradient Descent (SGD),
were combined with GridSearchCV, RandomizedSearchCV, Outof-Bag (OOB) validation, and SHAP values to raise both accuracy
and interpretability. After the best features were selected, thirteen
classifiers were trained and evaluated: LR, DT, RF, SVM,
Gaussian Naive Bayes (GNB), Multinomial Naive Bayes (MNB),
MLP, Linear Support Vector (Linear SV), K-Nearest Neighbors
(KNN), Gradient Boosting (GB), Extra Trees (ET), Bernoulli
Naive Bayes (BNB), and AdaBoost (AB). The models were trained
on datasets drawn from Software Development Versioning (SDV)
repositories that contain 1,000 log file entries. Across accuracy,
precision, recall, F1 score, and ROC AUC, ensemble models, and
RF in particular, gave the strongest results, reaching the highest
accuracy of 96.02% for Long Method and 92.63% for Large
Class. The findings were further confirmed by the Wilcoxon
signed-rank test. For practical use, the method was deployed
as a web application built with the React Native framework that
analyzes log data, flags smelly code segments, and reports their
severity to give developers actionable insight. The approach offers
a reliable and interpretable solution for Code Smell detection with
clear potential for integration into modern software development

End-to-End Explainable Stacked Ensemble Framework for Used Bike Price Prediction Using Real-World Marketplace Data

The significant research contribution of this study is the development of an explainable stacked ensemble framework for used-bike price prediction using real-world marketplace data from Bangladesh. The proposed framework integrates XGBoost, LightGBM, CatBoost, and HistGradientBoosting as base learners with Ridge regression as the meta-learner, supported by a comprehensive feature-engineering strategy that produces 28 predictors. Using 6,697 cleaned listings from Bikroy.com, the proposed model achieved an R^2 of 0.8759, RMSE of 0.2456, and MAE of 0.1712, outperforming all individually evaluated models. Furthermore, SHAP-based analysis provides interpretability by identifying engine capacity, vehicle age, and the brand–engine capacity interaction as major pricing factors. Finally, the complete framework was deployed as a 36 MB Android application using ONNX models, enabling on-device price prediction without cloud computation.

Pooled and Compromised: A Leakage Audit of a Popular MRI Benchmark

The Kaggle “Brain Tumor MRI Dataset” (Nick- parvar) is a saturated benchmark, with recent work reporting 97–99.9% accuracy under naive, patient-agnostic splits. It pools images from multiple sources without retaining patient identifiers, so near-duplicate slices from the same scan can land on both sides of a random split, inflating accuracy. We attempt to recover true patient identity for the Figshare-derived portion (glioma, meningioma, pituitary) by matching 7,200 pooled Kaggle images against the original 3,064 patient-labeled Figshare images, using a two-stage pipeline: normalized perceptual-hash retrieval with SSIM confirmation, then an ORB feature-matching fallback. This recovers a patient ID for 527 of 5,400 eligible images (9.8% coverage), a 49-fold improvement over a single-stage baseline (0.2%), though coverage remains far below what the dataset’s documented provenance implies, and only 52.2% class- label agreement among recovered images suggests many ORB- fallback matches are incorrect. We combine these identities with near-duplicate clustering over the remaining pool (24.7% multi- image clusters) to build a group-aware, leakage-corrected split covering all 7,200 images. Retraining three CNNs (ResNet50, EfficientNet-B0, DenseNet121) under both protocols shows the naive split overstates test accuracy by 0.93–1.59 points per architecture. A soft-voting ensemble reaches 98.06% accuracy under the corrected protocol, significantly better than the best single architecture, EfficientNet-B0, at 97.26% (McNemar’s test, p = 0.012), reversing a null ensembling result from a weaker, purely clustering-driven correction in an earlier pass. We release the corrected split, code, and all figures/tables for reuse.

Impact of Organic Fertilizer on Yield Soil and Health

Evaluation of Sustainable Nutrient Management: This paper provides a comprehensive review of structural, chemical, and biological transformations in soil health driven by organic fertilizers, biochar, cover crops, and bio-amendments such as Plant Growth-Promoting Rhizobacteria (PGPR) to counter intensive chemical fertilization.
Quantitative Analysis of Soil Properties: It synthesizes empirical datasets to deliver a quantitative breakdown of soil physical attributes (macro-aggregate stability, bulk density, porosity), chemical parameters (pH, electrical conductivity, sodium adsorption ratio, and soil organic matter), and biological/enzymatic activities like urease and phosphatase.
Crop Productivity and Quality Assessment: The research investigates physiological responses including chlorophyll accumulation (SPAD index), proline stress alleviation, and ultimate yield quality improvements such as protein, gluten, and total soluble solids across multiple crop species.
Integration of Advanced Technologies and Challenges: It discusses practical application constraints, nutrient release kinetics, and future technological integrations including GIS, Remote Sensing, and Precision Agriculture to optimize organic fertilizer management.

Bridging the Digital Divide in Secondary Education in Rural Bangladesh: Challenges, Impacts, and Policy Solutions.

This research provides a systematic review of the digital divide in secondary education in rural Bangladesh by analyzing findings from 45 studies. It brings together the major challenges, including poor internet infrastructure, socio-economic inequality, gender gaps, and limited digital skills. The study also examines how these issues affect students’ learning and identifies gaps between existing digital education policies and their implementation. Finally, it suggests practical directions for improving digital access, teacher capacity, and educational opportunities for rural students.

Retrospective Landmark Classification of Recorded Administrative Student Dropout Status from University Records

This study advances the methodology of educational data mining by establishing a leakage-controlled retrospective landmark evaluation framework that exposes the analytical boundaries of predicting student attrition from administrative records. Rather than treating repeated final administrative categories as dated withdrawal events, this research demonstrates why high random-split discrimination should not be mistaken for prospective intervention readiness.

A Comparative Analysis of COVID-19 Vaccine Side Effects vs. Environmental Allergies: A Machine Learning Approach to Symptom Differentiation

This research contributes a machine learning-based approach to distinguish COVID-19 vaccine-related side effects from environmental allergy symptoms. By analyzing demographic, vaccination, symptom, and environmental exposure factors, the study identifies patterns that may help improve symptom classification and support more informed clinical assessment. The research also provides a foundation for developing data-driven tools for differentiating vaccine-related reactions from common environmental allergies.

Machine Learning Techniques in Cybersecurity Threat Detection: A Systematic Literature Review

There is a systematic analysis of machine learning techniques for cybersecurity threat detection provided in this study. It contrasts conventional machine learning, deep learning, and ensemble methods in the most prominent categories of cyber threats, and reveals important challenges like adversarial attacks, imbalanced data sets, model interpretability, concept drift, and limited ability to generalize. Another key finding of the study is that it identifies several research gaps and opportunities that would guide future research in more adaptive, explainable, and region-specific cybersecurity solutions.

The Impact of Generative Pre-trained Transformer and Large Language Models on Students’ Academic Performance

This study combines machine learning and statistical analysis to identify key LLM usage behaviors associated with students’ perceived academic improvement. Random Forest achieved 82.6% accuracy and an F1-score of 0.895, while the findings highlight the importance of responsible and critical LLM use.

An Analysis Between Public and Private Universities

This review synthesizes existing evidence on inequalities between public and private higher education institutions. It examines disparities in funding, affordability, educational quality, infrastructure, student satisfaction, and graduate employability. The study develops a comparative understanding of these institutional differences while identifying key methodological and empirical gaps in the existing literature. The findings provide a basis for informed policy development and future research toward greater equity in higher education.

An Investigative Survey on Depression Detection from Social Media Data Using Machine Learning and Deep Learning Models

This study reviews different machine learning and deep learning methods used to detect depression from social media data. It compares commonly used models and techniques and discusses their strengths, limitations, and challenges. The study also highlights possible directions for improving the accuracy and reliability of depression detection systems.

Principal Component Analysis Implementation on Machine Learning in Diabetes Classification

Stroke Prediction in Young Adults (Age 18-45): A Machine Learning-Based Preventive Approach

Impact of climate change on agriculture in Bangladesh

Kidney Disease Detection Using Machine Learning with Explainable Artificial Intelligence

This research provides a comprehensive systematic review of machine learning-based approaches for Chronic Kidney Disease (CKD) detection with a focus on Explainable Artificial Intelligence (XAI). The study identifies the effectiveness of widely used ML models, including Random Forest and XGBoost, and analyzes how explanation techniques such as SHAP and LIME improve model transparency. Additionally, it highlights key research challenges, including limited dataset diversity, insufficient clinical validation, and the need for interpretable AI solutions in healthcare. The findings provide valuable insights for developing reliable, transparent, and clinically applicable CKD detection systems.

Urban and Rural Areas Studies Difference Using Machine Learning

This research contributes a machine learning-based approach for identifying and analyzing the differences between urban and rural areas using socioeconomic and demographic features. The study compares multiple machine learning models and demonstrates that Random Forest can effectively classify urban and rural areas, achieving an accuracy of 94.20%. The findings provide data-driven insights that can support better planning, resource allocation and strategies for reducing urban–rural disparities.

Impact of Social Media on Mental Health

This research study that how affecting on our mental health in every possible way. we identifies some risk factors and highlights how machine learning and deep learning techniques can detect mental heatlh issue from social media data.

Educational Differences Between Urban and Rural Schools: Challenges and Opportunities in Rural and Urban Education

This study looks at the differences between rural and urban education in Bangladesh using survey data from 100 students. It focuses on factors such as teachers, internet access, transportation, and school facilities. Three machine learning models were also tested, with Random Forest giving the best results.

Impact of social media on students academic performance

This study examines how social media use affects students’ academic performance, study habits, and concentration, providing insights for promoting balanced and effective social media use.

Predicting Dinosaur Coloration from Fossilized Melanosome Data Using Classical Machine Learning

This study establishes a reproducible classical machine-learning baseline for dinosaur coloration reconstruction from fossilized melanosomes, while systematically evaluating cross-clade generalization. By comparing QDA, SVM, and Random Forest on theropod and non-theropod taxa, the study demonstrates that high-confidence predictions for phylogenetically distant specimens reflect distribution shift and morphological overlap with the avian training data rather than reliable biological coloration. These findings highlight the limitations of transferring bird-trained models across dinosaur clades and motivate more taxonomically broad and taphonomically informed approaches to paleocolor reconstruction

Factors Contributing to University Student Dropout Using Machine Learning

Explainable Machine Learning framework for Antibiotic Resistance Prediction in E. coli UTIs: Clinical Insights for Bangladesh

Urinary tract infections caused by Escherichia coli are among the most common bacterial infections in clinical practice. Increasing antibiotic resistance has made empirical treatment increasingly challenging particularly in resource-limited settings such as Bangladesh. This study proposes an explainable machine learning framework for predicting antibiotic resistance in E. coli UTI patients using routine clinical and laboratory data. Two publicly available datasets were merged to create a
dataset of 10,710 patient records with 32 processed features. Five machine learning models including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine and XGBoost were evaluated using 5-fold stratified cross-validation across five antibiotic resistance targets including Ciprofloxacin, Gentamicin, Amoxicillin-Clavulanic Acid, Cefotaxime/Ceftriaxone and Amoxicillin/Ampicillin. SHAP and LIME were employed to provide global and patient level model explanations. SVM achieved the best overall performance for AMC, CTX/CRO and AMX/AMP with the highest AUC of 0.586 for CTX/CRO, while XGBoost obtained the highest accuracy for CIP (83.87%) and GEN (78.33%) but detected almost no resistant cases at those operating points. SHAP identified age, diabetes status, body temperature, hospital admission history, hypertension and the Neutrophil-to Lymphocyte Ratio as the most influential predictors, consistent with risk factors reported in Bangladeshi clinical studies. These findings demonstrate that explainable machine learning using routinely available clinical data can support early antibiotic resistance prediction and assist empirical antibiotic selection in resource-limited healthcare settings.

Predicting E-Learning Efficacy and Gap Closure in Post-Pandemic Higher Education Using Machine Learning

The COVID-19 pandemic shifted higher education from traditional in-person classrooms toward flexible digital delivery. While emergency remote teaching preserved instructional continuity during lockdowns, it also widened academic disparities. This study examines whether online education has since closed the learning gap it helped create. We analyzed 509 survey responses collected across the eight administrative divisions of Bangladesh, yielding 284 unique response profiles after duplicate removal. Using a quantitative supervised machine learning framework, we evaluated seven socio-technical predictors: administrative division, internet access quality, device availability, teacher online class hours, student engagement, community readiness, and institutional support. Random Forest, K-Nearest Neighbors, and Support Vector Machine classifiers were tuned by grid search with 5-fold cross-validation on the training partition and assessed on a held-out test set. Random Forest achieved the strongest holdout performance, with a test accuracy of 84.21% and a weighted F1-score of 0.83, while SVM attained the highest cross-validation accuracy at 83.29%. Feature importance analysis identified Institutional Support as the dominant contributor at 0.54, followed by regional context at 0.12. Across the evaluated classifiers, the surveyed cohort predominantly reflected a perception that the educational gap was only partially closed. These findings indicate that although online platforms have improved flexibility, access, and instructional continuity, virtual tools alone cannot replicate the collaborative presence of physical classrooms. Digital learning therefore works best as a complementary component of a balanced hybrid model rather than a full replacement for traditional instruction.

A Machine Learning-Based Comparative Analysis of Rural and Urban Educational Inequality in Bangladesh

Educational inequality between rural and urban areas remains a significant challenge in Bangladesh, where differences in infrastructure, technology, and academic support shape students’ learning outcomes. This study investigates these disparities using survey data collected from 145 respondents, organized into separate primary and high school datasets and analyzed using descriptive statistics and supervised machine learning. The analysis focuses on educational resources, digital facilities, internet access, private tutoring, teacher support, and academic performance. Four classification models were com pared: Logistic Regression, Support Vector Machine (SVM), Random Forest, and Extreme Gradient Boosting (XGBoost), each trained to classify rural and urban educational backgrounds and evaluated using accuracy, precision, recall, F1-score, and five fold cross-validation. The results reveal substantial rural–urban differences in educational resources, with urban students reporting higher internet availability, greater access to digital devices, more computer laboratory facilities, and higher rates of private tutoring, and better facilities associated with improved academic performance. Logistic Regression achieved the highest accuracy, reaching 79% for the primary school dataset and 83% for the high school dataset. The findings highlight greater inequality at the high school level and demonstrate the potential of machine learning to identify educational disparities, supporting the need for improved educational infrastructure and digital accessibility in Bangladesh.

Mental Health Classification from Social Media Data in Banglish using Transformer-Based Multi-Task Learning and Explainable AI

1. We have released a public Banglish mental health dataset of 56,522 samples.
2. We have presented a multi-task architecture named BanglaBERT-MTL-AP.
3. We have introduced a three-layer hierarchy routing module and Grad*Input based XAI system for the interpretability.

IETS-Stroke: An Imbalance-Aware, Explainable, and Threshold-Tuned Stacking Framework for Stroke Risk Screening with Cross-Dataset Validation

This research proposes IETS-Stroke, an imbalance-aware, explainable, and threshold-tuned stacking framework for stroke risk screening using public-health datasets. The framework combines imbalance handling, stacking ensemble learning, threshold optimisation, SHAP explainability, and BRFSS-to-NHANES external validation. It can be used as a screening-support approach to identify individuals who may require further clinical assessment, rather than as a diagnostic system.

Confidence-Aware Retrieval-Augmented Generation with Abstention for Cyber Threat Intelligence

Retrieval-Augmented Generation (RAG) grounds
language models in external knowledge, which makes it a
natural tool for Cyber Threat Intelligence (CTI). It has one
dangerous habit, however: it answers every question, even when
the retrieved evidence does not support an answer. The result
is hallucinated vulnerability details that can mislead a security
analyst. We present a confidence-aware RAG framework for
retrieving CVE-based threat intelligence. The framework adds a
tunable abstention step. When its confidence in the retrieval is too
low, the system declines to answer instead of guessing. We measure
this confidence at the retrieval stage from two simple signals: how
similar the best-matching record is to the query, and how far
ahead it is of the second-best record. The measure needs no access
to the language model’s internals. We evaluate three systems,
Naive RAG, Hybrid RAG, and our Confidence-Aware RAG, on
200 keyword-based queries over the 2023 National Vulnerability
Database (30,932 CVE records). At a confidence threshold of
0.20, our system answers with 81.3% precision, against 70.0%
for a baseline that always answers, and it lowers the Expected
Calibration Error from 0.309 to 0.259. In short, for security-
critical CTI work, a few trustworthy answers are worth more
than many unreliable ones.

Calibrated and Explainable Credit Risk Assessment: Integrating SHAP Analysis with Feature Stability Validation

Machine learning has substantially improved the predictive performance of credit risk assessment, yet high pre- dictive accuracy alone is insufficient for deployment in regu- lated lending, where reliable probability estimates, interpretable decisions, and consistent explanations are equally important. Existing studies typically evaluate these properties independently, leaving uncertainty about whether a model can satisfy all three simultaneously. This paper proposes an integrated evaluation framework that jointly assesses discrimination, probability cali- bration, and explanation stability with statistical confidence. Us- ing the German Credit dataset, we compare Logistic Regression, Random Forest, and LightGBM. Model performance is evaluated using ROC-AUC, the Brier score, Expected Calibration Error (ECE), bootstrap confidence intervals, and paired significance tests, while TreeSHAP explanations are validated across five cross-validation folds using feature-wise coefficients of variation. Experimental results show that LightGBM achieves the best overall discrimination and calibration while maintaining stable explanations for its most influential risk factors. However, its performance advantage over Random Forest is not statistically significant. We further demonstrate that synthetic oversampling improves recall but consistently degrades probability calibration, whereas cost-sensitive threshold optimization achieves compara- ble gains without distorting predicted probabilities. In addition, a fairness audit reveals demographic disparities that are not reflected in feature attribution alone. These findings highlight that discrimination, calibration, explanation stability, and fairness should be evaluated together, providing a more reliable basis for selecting and validating credit risk models for practical deployment. Index Ter

Agentic AI in Software Engineering: Evidence from Production Software and Future Directions for Developing Countries

The significant research contribution is the empirical evaluation of agentic AI coding tools in a production-style software environment. The study uses 160 tasks performed by 16 professional .NET developers on the nopCommerce platform to quantify success, task-complexity effects, human intervention, token exhaustion, and failure patterns. It demonstrates that task complexity significantly affects agentic AI performance and provides practical recommendations for human-in-the-loop adoption of agentic AI in Bangladesh and similar developing-country software environments

Retrieval-Augmented Legal Question Answering Using Hybrid Semantic Search

1. A full ablation of sparse, dense, fusion-based and document-aware retrieval on the complete benchmark, scored with a character-level metric.
2. A weighted fusion scheme that repairs the failure of equal-weight reciprocal rank fusion, and a sparse backend (bm25s) that cuts BM25 scoring time by three orders of
magnitude.
3. A two-stage document-aware retriever that narrows the search to the correct document before re-ranking, the strongest method we tested.
4. A citation-grounded generation stage whose citations are scored against gold spans, showing that better retrieval yields better-attributed answers even though the model tends to under-cite.

A Multi-Dimensional Performance Assessment of Rooftop Agrivoltaic Systems for Sustainable Urban Development

This research contributes to the development of agrivoltaic systems by investigating the dual-use of agricultural land for photovoltaic electricity generation and crop cultivation, with emphasis on optimizing PV installation conditions to maintain energy production while supporting agricultural productivity. The findings provide practical insights for sustainable land use, renewable energy generation, and climate-resilient agriculture, particularly in regions such as Bangladesh.

Experimental Investigation of Rooftop Solar PV Deployment for Passive Building Cooling and Sustainable Electricity Generation

This study demonstrates the dual functionality of rooftop photovoltaic (PV) systems as both renewable energy generators and passive thermal-shielding elements. The key contribution is the experimental quantification of rooftop cooling using thermal imaging across three different PV installation heights, showing an average 25.97 °C (46.14%) reduction in rooftop surface temperature. The findings provide practical evidence that strategic PV deployment can simultaneously reduce building heat gain, potentially lower cooling demand, and support clean energy generation, offering a practical approach for energy-efficient and climate-resilient urban buildings.

The Adoption of Internet of Things in Higher Education: Opportunities, Challenges, the Role of vision 2030 in Saudi Arabia

Empirical baseline — Establishes that 65% of Saudi universities have deployed some IoT, with adoption skewed toward well-resourced institutions, giving policymakers a concrete starting point.
Integrated theoretical lens — Applies TAM and DOI together to explain why adoption succeeds or stalls in this specific national context, not just whether it does.
Three-factor barrier model—Identifies and categorizes the interplay of technical, economic, and cultural-organizational barriers as a coherent obstacle framework specific to Saudi Arabia, rather than treating each in isolation.
Vision 2030 alignment — Positions IoT as a measurable instrument of national policy, connecting technology adoption research to a live strategic agenda—giving the findings direct policy relevance beyond academic circles.
Actionable recommendations—Translates findings into three concrete policy levers: infrastructure investment, training programs, and regulatory frameworks.

Explainable Multi-Class Electricity Theft Detection on the TDD2022 Benchmark

This work presents the first explainable, per-attack-type analysis of electricity theft detection on the TDD2022 benchmark. Ten classifiers are evaluated under a frozen, temporally ordered, per-building-type split that prevents leakage between training and test sets. Rather than reporting aggregate accuracy alone, the study reveals that detection difficulty varies dramatically by attack type from perfect detection to near-complete evasion and uses SHAP to attribute why each attack is caught or missed to specific consumption features. A McNemar’s test across ten models, including modern gradient-boosting methods, demonstrates that performance is bounded by the feature representation rather than model choice, motivating sequence-aware detection as future work.

A Human-Centred Critical AI Literacy Framework for Higher Education

This research contributes a novel Seven-Stage Critical AI Literacy Framework that shifts higher education responses to GenAI from surveillance and compliance towards the development of evaluative judgement, ethical reasoning and reflective accountability. The framework provides a structured, assessment-oriented approach for critically evaluating AI-generated outputs across seven dimensions and makes human judgement and intellectual accountability visible and assessable. It offers a practical foundation for assessment redesign, curriculum development and responsible AI integration across disciplines.

Cross-Lingual and Cross-Domain Transfer Learning for Digital-Arrest Scam Detection via Transformer Warm-Starting with Coercion-Stage Explainability

Real-Time Food Spoilage Detection Using Multi-Sensor Gas Analytics and IoT-Based Monitoring

This study introduced an IoT-enabled real-time food rotting detection architecture powered by dynamic edge computing and multi-sensor gas analytics. The system effectively adjusted its categorization limits (Td) to real-time microclimate fluctuations by merging MQ-series gas data into a single index (ΦGas) and using a continuous multivariate contextual compensation function. The system successfully neutralizes baseline drift noise caused by environmental humidity and temperature fluctuations while achieving an absolute classification accuracy of 100% across transitional decomposition phases, according to experimental validation using telemetry streams.

Should Bangladesh Adopt Age Verification Laws? A Socio-Legal Analysis of Surveillance, Privacy, and Public Trust

This study examines the implications of age-verification (AV) laws for Bangladesh, where digital governance is evolving but public trust in data handling remains low.
The paper recommends rights-based alternatives, including digital literacy initiatives, parental empowerment, and privacy-preserving age assurance, as more context-appropriate strategies for protecting children online.

Development and Preliminary Evaluation of Banana Stem Fiber-Based Biocomposite Using Natural Okra Mucilage and Rice Gum Binders

This study develops a novel bio-composite from banana stem fiber and coconut particulate filler by the use of two food grade natural binders such as okra mucilage and rice gum paste as fully plant based alternatives to synthetic resins. Through direct qualitative comparison, okra mucilage was found superior to rice gum paste in rigidity, surface quality and fiber binder adhesion. It’s also retaining strong adhesion under water and oil exposure. The work establishes a locally sourced low cost and biodegradable composite system aligned with circular economy principles and UN SDGs 9, 12, and 13 laying the foundation. This is a continuous process, the strength of the newly formed composite is yet to confirm, however the composite shows better adhesion.

A Dual-Backbone Hybrid Deep Learning Framework with Squeeze-and-Excitation Attention and Grad-CAM Explainability for Multi-Class Brain Tumor MRI Classification

Accurate brain tumor classification from magnetic
resonance imaging (MRI) is essential for treatment planning,
yet manual interpretation remains time-consuming and subject to inter-observer variability. This paper proposes a dual-
backbone convolutional neural network that fuses EfficientNetB3
and MobileNetV2 features through a Squeeze-and-Excitation
(SE) channel-attention block for classifying glioma, meningioma,
pituitary tumor, and no-tumor MRI scans. The preprocessing
pipeline combines Gaussian–bilateral denoising with contrast-
limited adaptive histogram equalization (CLAHE) to suppress
noise while preserving diagnostically relevant contrast. The model
was trained progressively through frozen-backbone warm-up,
partial fine-tuning, and full end-to-end optimization with cosine
learning-rate scheduling on 7,153 MRI images. The proposed
network achieved 99.72% validation accuracy and 99.16% test
accuracy, with macro precision, recall, F1-score, and ROC-AUC
values of 0.9914, 0.9917, 0.9915, and 0.9997, respectively. Ablation
analysis showed that direct backbone fusion without SE attention
reached only 95.35% accuracy, whereas the complete attention-
enhanced model achieved 99.16%, confirming the importance
of adaptive channel recalibration. Grad-CAM was used to
visualize class-relevant tumor regions and improve prediction
transparency. These findings demonstrate that the proposed
attention-guided dual-backbone framework provides accurate,
interpretable, and computationally practical brain tumor classification for clinical decision-support applications.

A Cross-Environment Benchmarking Study of Quantum Machine Learning for Wi-Fi Deauthentication Attack Detection Using Public Datasets and Real-World Captures

This study provides a leakage-resistant, cross-environment comparison of quantum and classical models for Wi-Fi deauthentication detection using public and real-world data, including target-domain adaptation and actual quantum hardware evaluation.

Real-Time Object Classification in Autonomous Driving: A Lightweight Residual CNN Approach on Unified BDD100K–COCO Data

Unified Multi-Dataset Super-Taxonomy: A novel three-class (Vehicle, Pedestrian, Obstacle) crop-based data unification pipeline integrating BDD100K and MS COCO 2017, yielding 58,831 annotated ROI samples with a consistent label space for real-time classification.
Lightweight Residual CNN Architecture: A custom four-stage residual backbone (channels 32→512) with global average pooling and dropout regularisation, specifically designed for resource-constrained on-board automotive hardware.
Speed–Accuracy Optimisation: Achievement of 215.5 FPS inference throughput via 224×224 input resolution and FP16 mixed-precision acceleration while maintaining 93% mean classification accuracy and per-class F1-scores of 0.94/0.94/0.91.

Analysis of Data Privacy, Security, and User Perception in VPN Applications

This paper is significant because it examines data privacy, security, and user perception in VPN applications, which are increasingly important for protecting users’ online activities. It evaluates how effectively VPNs safeguard sensitive data and explores users’ trust, awareness, and concerns regarding VPN security and privacy. The findings can help VPN developers, cybersecurity professionals, and users identify security weaknesses, improve privacy protection, and make more informed decisions when selecting VPN services.

Road Surface Defects and Utilities Detection using YOLO26 with Explainable AI

1. We conduct a comprehensive comparative evaluation of six YOLO models (YOLOv7, YOLOv8, YOLOv9, YOLOv10, YOLOv11, and YOLO26) for multi-class road surface anomaly detection.

2. We design a multi-class detector that is capable of identifying potholes, drain holes and sewer covers correctly based on the images of the roads.

3. We use explainable AI, grounded on Grad-CAM visualization, to increase model transparency, and to gain insight into spatial attention behavior on model.

4. We analyze the models using conventional object detection metrics and show that YOLO26 has the greatest overall detection performance on the studied dataset.

Enhancing Power Factor of EV Chargers: Sustainability & Future Prospects Using Capacitor Banks & Synchronous Condensers

The widespread deployment of wireless electric vehicle charging systems poses major issues for distribution networks such as high reactive power demand, voltage fluctuation, voltage imbalance, power factor issues and harmonic distortion. This study suggests a hybrid reactive power compensation method, which is a switched capacitor bank for steady state support and synchronous condenser for fast dynamic compensation. The charging-system model developed in the project contains the 3-phase grid source, measurement and control units, AC-DC converter, lithium-ion battery and proposed compensation devices. The results of the study indicate that the hybrid system is able to effectively control the sudden changes in reactive power during multi-step wireless charging scenarios. After a large short-term reactive power surge, the power factor recovers to 0.999 in about 4ms and settles at 0.9914. The system also ensures stable battery charging, regulated terminal voltage and better supply current–voltage alignment, whilst retaining the same active power transfer. The proposed method is less complex in implementation, easier to realize and less switching stress than the active power-factor-correction rectifier. Thus, the hybrid configuration offers a viable, robust and scalable power quality enhancement solution as well as wireless electric vehicle charging infrastructure.

Electric Vehicle Inverter Topologies and Total Harmonic Distortion Mitigation: A Comprehensive Study of Recent Advances and Future Trends

The motor performance, powertrain efficiency, driving range and power quality of an electric vehicle (EV) are greatly affected by the electric vehicle (EV) traction inverter. This paper gives a detailed survey of the inverter topologies and total harmonic distortion (THD) reduction methods published in the literature during the last few years. Conventional two-level voltage-source inverters are compared to multilevel voltage-source inverters, such as neutral-point-clamped, T-type, flying-capacitor, cascaded H-bridge, reduced-switch-count and switched-capacitor topologies, with focus on 400 V and 800 V EV architectures. Other modulation and control techniques like sinusoidal pulse-width modulation, space-vector modulation, selective harmonic elimination, nearest-level control and model-predictive control are also assessed. The results show that the three-level NPC and T-type inverter are very suitable for 800 V systems, which can significantly reduce THD and inverter losses, compared to two-level systems. Moreover, switching performance, efficiency, and power density are enhanced even more by the use of wide-bandgap semiconductors, such as SiC MOSFETs and GaN HEMTs. But harmonic performance is also dependent on the variation in the DC link voltage, dead time effects, motor parameters and operating conditions. Issues of adaptive real-time control, reliability-aware design, fault-tolerant operation, and integrated charger–traction inverter systems should be addressed in future research.

Performance and Degradation of Advanced Mobile Batteries: Future Prospects for Sustainable Battery Development

This research focuses on comparative analysis of battery health degradation in Lithium-Ion (Li-Ion), Lithium-Polymer (Li-Po) and Silicon-Carbon (Si/C) battery chemistries for future mobile energy storage applications. The key degradation mechanisms, cycle durability, capacity retention, energy density, thermal performance and barriers to large scale adoption are explored. Although Li-Ion does not have the lowest weight or smallest form factor, it is the industry standard because of the existing manufacturing infrastructure, understanding of its degradation characteristics, and a typical cycle life of 1,000 to 2,000 cycles. Although Li-Po batteries offer versatility of shape, low weight and enhanced leakage resistance, they have limited application for high voltage because of the degradation of polymer electrolyte and the instability at the interfaces. While silicon based batteries (Si/C) have a much higher theoretical capacity and superior practical energy density, silicon expansion of greater than 300% causes particle fragmentation, loss of electrical contact, multiple formation of solid electrolyte interphase (SEI) and rapid capacity loss. In the real world, Samsung’s Si/C battery illustrates the importance of the capacity vs mechanical expansion vs cycle life vs reliability balance. Results show that the choice of an optimal battery technology is very application-specific. Advanced Si/C nanostructure engineering, prelithiation strategies, durable binder materials, next generation polymer electrolytes, scalable manufacturing processes, and AI-based battery management systems for enhanced lifetime performance and degradation prediction are the future research priorities.

Comparative Analysis of a YOLO-Based Model, Custom CNN, and ResNet18 for Facial Emotion Recognition

Facial emotion recognition remains challenging under unconstrained conditions because of variations in pose,
illumination, occlusion, image quality, and class distribution. This
paper compares three deep learning architectures for eight-class
facial emotion recognition: YOLOv8m, a custom convolutional
neural network, and ImageNet-pretrained ResNet18. All models
were trained and evaluated using the same prepared AffectNet
partitions containing 25,262 images. Model-specific preprocessing
and augmentation were applied while preserving a common
training, validation, and test split. YOLOv8m achieved a mean
average precision of 0.8328, while the custom CNN and ResNet18
obtained test accuracies of 56.32% and 71.32%, respectively.
Class-level analysis showed that Fear produced the strongest
YOLOv8m performance, whereas Disgust remained the most
difficult emotion, partly because of its lower sample count
and visual similarity to other negative expressions. The best
YOLOv8m checkpoint was integrated into a Flask-based backend
and deployed through a web application built with HTML, CSS,
and JavaScript, and an Android application developed using Java
in Android Studio. Direct numerical comparison between mAP
and classification accuracy remains limited because the evaluated
models use different output formulations.

Performance Analysis of Mobile Phone Chargers Under Different Charging Loads: Power Quality for Sustainable Charging and Future Prospects

The low-power switched-mode chargers used to power mobile phones can cause power quality issues when installed in large numbers. In this study, there were five anonymized mobile phones (M1-M5) and three anonymized adapters (Charger 1: 18W, Charger 2: 33W, Charger 3: 45W). The number of 15 operating points were analyzed based on the input current total harmonic distortion (〖THD〗_i), power factor (PF), RMS current and active power. Mean 〖THD〗_iincreased from 119.02% for Charger 1 to 148.08% for Charger 2 and 162.98% for Charger 3, whereas mean PF decreased from 0.6146 to 0.5526 and 0.5070, respectively. In all observations, there was a strong inverse relationship between 〖THD〗_i and PF (r = -0.974, p < 0.001). The results show a power quality problem while the phone dependent operation and part load operation, but lack of topology and protocol information makes causal generalization and compliance classification impossible. The results inspire the load-aware design of chargers, low loss current shaping, clear power-quality reporting, and standardized full cycle measurements for sustainable charging.

Dynamic Traffic Light Control System Using Wireless Sensor Network

We have calculated the vehicle priority based on vehicle
weight, length, and carbon emission rate. And also given
higher priority to those vehicles which have higher passenger
capacity. Lengthy vehicles for instance buses are given high
priority because of their higher passenger capacity. They can
accommodate more passengers than small vehicles like cars,
and bikes.
• We’ve generated a formula to prioritize the length considering the vehicle priority and the number of vehicles
up to a certain distance.
• The green signal timing of each lane will be calculated
based on the required time for passing the different types
of vehicles in each lane. Then using the wireless sensor,
control the duration of the traffic light and determine the
phase.
• Our proposed DTLCS also handles emergency vehicles.
When an EV (emergency vehicle) enters a lane, we have
developed a method to pass the EV as soon as possible
by changing the lane priority. The system has the ability
to resume the previous operation in an efficient way
implemented in PTV-VISSIM simulator

Quantifying the Saline Penalty: A Composite Degradation Framework for Techno-Economic Assessment of Floating PV on Coastal Shrimp Ghers in Bangladesh

Floating solar photovoltaics (FPV) have been deployed on freshwater ponds in Bangladesh, but their potential on the saline shrimp ghers of the coastal southwest remains unexplored. This paper proposes the Saline Penalty (SP); a transferable decision-support metric defined as the percentage increase in Levelized Cost of Energy (LCOE) due to saline-induced degradation. Using PVsyst simulation for a 101 kWp FPV system on a representative shrimp gher supplying Mostafa Organic Shrimp Products Ltd. (MOSPL) in Shyamnagar, Satkhira, Bangladesh, we develop a composite degradation framework informed by physically based degradation mechanisms from offshore PV literature. The framework incorporates salt deposition (0.05–0.13% annual acceleration), corrosion (transmittance loss from 91.46% to 70.35%), and biofouling using a multiplicative formulation. Uncertainty analysis via Monte Carlo simulation (10,000 iterations) quantifies the probability of SP exceeding the viability threshold. Results show the Saline Penalty ranges from 48.4% (optimistic, with mitigation) to 101.7% (pessimistic, without mitigation). Sensitivity analysis identifies viability thresholds: projects become economically unviable when SP exceeds ~60% (degradation >1.1–1.2% annually or CAPEX premium >17–18%). Monte Carlo simulation reveals a 34.4% probability of SP > 60% in the optimistic scenario and 100% in the pessimistic scenario. CO₂ avoided ranges from 95.7 to 116.3 tonnes annually. Based on the literature reviewed, this appears to be among the first studies to quantify the saline penalty for FPV in Bangladesh’s coastal ghers, providing a reusable framework for decision-makers evaluating FPV deployment in saline environments which is classified into four categories—Low (<20%), Moderate (20–40%), High (40–60%), and Critical (≥60%); enabling rapid decision-making.

Beyond Whole-Slice: Comparing Encoder Families and Patch-Based Pipelines for Pancreas Segmentation

(1) Comprehensive encoder benchmark: Systematically compares representative CNN, Transformer, and hybrid CNN–Transformer encoders for pancreas segmentation under a unified experimental setting to identify the relative strengths of different feature-extraction paradigms.

(2) Progressive segmentation-pipeline evaluation: Evaluates four segmentation pipelines designed to address key challenges in pancreas CT segmentation, including severe foreground–background imbalance, low boundary contrast, and substantial anatomical variability.

(3) Rigorous performance–efficiency analysis: Evaluates all architecture–pipeline combinations using identical five-fold cross-validation and analyzes segmentation performance alongside model size and inference time, identifying the combination that provides the best balance between accuracy and computational efficiency.

Air Quality Index Forecasting in Dhaka Using Deep Learning: A comparative study of LSTM, GRU, and BiLSTM+Conv1D

This study presents a comparative deep-learning framework for multi-step Air Quality Index (AQI) forecasting in Dhaka using LSTM, GRU, and BiLSTM+Conv1D models trained on a 25-year hourly time-series dataset. The research systematically evaluates the impact of 6 h, 12 h, and 24 h historical input windows on forecasting accuracy and demonstrates that the hybrid BiLSTM+Conv1D model achieves the best overall performance (R² = 0.9948, RMSE = 1.5773, MAE = 1.1358). The work highlights the joint importance of model architecture and temporal context for accurate urban AQI prediction and supports the development of intelligent environmental monitoring systems.

YOLO-HVConv: A Horizontally-Vertically Aware YOLO for Road Damage Detection

Road damage detection is an important component
of the intelligent transportation system, which helps in timely
maintenance of roads and enhances transportation safety. How
ever, accurate detection of road damages is still a challenge due
to the presence of small, thin and elongated crack structures,
different damage scales and class imbalance. To address these
issues, this paper proposes an enhanced YOLOv8s-based frame
work for road damage detection by integrating the Horizontal
and Vertical Convolution (HVConv) modules at multiple stages
of feature fusion in the neck. To better model thin and elongated
road damage patterns, we incorporate HVConv into the P3 and
P4 feature levels in the proposed architecture while maintaining
the lightweight property of the original detector. An experiment
was carried out on a combined dataset of the China Drone and
China MotorBike subsets of the RDD2022 dataset. Comparative
evaluations were conducted against the baseline YOLOv8s model
and other HVConv configurations. The proposed YOLO-HVConv
model obtained the best overall performance (i.e., 82.57% preci
sion, 79.89% recall, 83.12% mAP@50, and 51.88% mAP@50:95).
Moreover, the model was more effective in detecting small
and medium road damages, with AP values of 55.71% and
40.57% respectively. The experimental results demonstrate that
the suggested integration method of multi-scale HVConv can
enhance the feature representation for road damage detection
while maintaining computational efficiency, making it suitable
for practical intelligent road monitoring applications.

HybridGuard: A Graph-Temporal Fusion Architecture for Real-Time IoT Intrusion Detection on Identity-Less Flow Data

This work makes two contributions to graph-based intrusion detection for identity-less IoT flow data. First, it introduces a proxy session graph construction method that enables relational modeling on datasets like CIC-IoT-2023, where conventional endpoint-identity-based graphs (IP/MAC) cannot be built. Second, through rigorous multi-seed statistical validation, it demonstrates that graph structure alone is insufficient in this setting — the GNN-only variant underperforms a simple tabular baseline on both tasks — while a gated fusion of spatial (GAT) and temporal (Transformer) representations yields a statistically significant improvement over Random Forest on binary detection (p = 0.0001), with real-time-feasible latency (3.66 ms/window). This finding reframes the value proposition of graph-based IDS: not as a standalone improvement, but as a component that becomes effective only when fused with temporal modeling.

SCAPS-1D Modeling and Multi-Parameter Optimization of a CABTSe/MASnI3 DoubleAbsorber Lead-Free Solar Cell

This study reports the first systematic, multi-parameter co-optimization of a lead-free double-absorber solar cell (DASC), jointly tuning ETL selection, absorber thickness, acceptor doping, temperature, and back-contact work function within a single CABTSe (CuAgBeSnSe4)/MASnI3 device architecture—an integrated approach that prior DASC studies had not addressed, since earlier work optimized these parameters individually rather than in combination. Using SCAPS-1D numerical simulation, we screened six ETL materials (WS2, SnS2, ZnSe, ZnO, PCBM, and TiO2) against a Zn3P2 hole transport layer and identified SnS2 as the optimal ETL,the optimal FTO/SnS2/CABTSe/MASnI3/Zn3P2/Au architecture, achieving 32.53% power conversion efficiency (PCE), while acceptor-doping optimization alone yielded 38.26% PCE, providing practical design guidelines for high-efficiency, environmentally friendly next-generation photovoltaic devices.

A Super-Convergent Knowledge Distillation Framework for Domain-Specific Text Classification

Large language models have pushed the recent-most
in Natural Language Processing (NLP), but are often too high
cost to deploy in safety-critical, domain-specific use cases such
as biomedical text processing, where efficiency is essential. This paper proposes a knowledge distillation framework for learning
lightweight and effective models. The framework rests on two
key methodological pillars, firstly a multi-teacher ensemble that
supplies diverse, strong knowledge representations to the student,
and secondly an adaptive, multi-component loss function that
shifts focus among several learning signals during training. The
overall objective is to achieve what we call super-convergence,
a setting where a compact student model retains much of its
teachers’ capability while remaining highly competitive with a
much larger ensemble. Experiments on a biomedical sentence
classification task (PubMed 20k RCT) show that a distilled
DistilBERT student with about 66M parameters achieves 0.8659
accuracy and 0.8623 weighted F1, coming within roughly 0.9
F1 points of a ∼287M-parameter heterogeneous BERT-family
ensemble. In practice, this corresponds to a more than 4.3×
reduction in parameter count and substantially faster inference,
with only a marginal loss in predictive performance. These
results highlight a practical direction for building resource
conscious yet high-performing AI systems in specialized domains.

Privacy-Preserving Sentiment Analysis using PATE-based Federated Learning

Due to the proliferation of user-generated text data
on personal devices, sentiment analysis has become more important for understanding public opinion, business performance,
and mental health indicators. However, existing methods require
the raw private text data to be sent to a central server, a
practice in direct violation of data protection policies such as
GDPR and HIPAA. In this paper, a new framework called PATE-FL, integrating Federated Learning and the Private Aggregation
of Teacher Ensembles, is introduced to achieve strong (ϵ, δ)-
differential privacy and maintain competitive task performance.
Eight teacher models are trained on three heterogeneous clients, with all inter-client communications being AES-GCM encrypted. Through knowledge distillation, a single student model is then trained on noisy soft label data generated by aggregating the votes of all teacher models on approximately 3,500 unlabeled public text samples, thus eliminating any direct exposure of the final model to raw private text data. Performance evaluation on a balanced 50,000-sample text corpus results in 85.94% ± 0.30% accuracy and macro F1, achieved at a strong level of privacy characterized by ϵ ≈ 3.5 and δ = 10^−5 . These results demonstrate a significant advancement of existing privacy-aware methods in terms of performance and privacy guarantees.

A Hydro-Mechanical Framework for Slope Mounted Agrivoltaic Infrastructure: A RUSLE Based Framework for Dripline Scour Mitigation and Foundation Stability

Bangladesh’s renewable energy roadmap targets 60,000 MW by 2041, necessitating solar deployment on steep,erosion-prone slopes in the Chittagong Hill Tracts; yet the geotechnical implications of hillside PV infrastructure remain critically understudied. This paper introduces a hydro mechanical framework that couples RUSLE-calibrated soil loss prediction with a geometric concentration factor and infinite slope stability analysis to evaluate foundation resilience on a 21° engineered terrace at Alutila, Khagrachhari. Rainfall intercepted by 3m-wide PV panels concentrates at the 0.1m-wide drip edge, creating a 30x concentration factor (CF = 3.0/0.1 = 30) that accelerates localized scour directly around pile foundations. Under bare slope conditions, erosion of 86.62 t/ha/yr translates to 185.6 mm/yr of localized scour, reducing pile embedment from 1.5m to 1.0m over five monsoon cycles and increasing the overturning lever arm by 33.3%. The proposed coupled PV+Vetiver system reduces scour to 1.18
mm/yr (99.4% reduction), preserving 99.3% of design embedment. Infinite slope stability analysis shows the Factor of Safety (FoS) recovering from 0.98 (UNSTABLE) under bare conditions to 1.76 (STABLE) within two years of Vetiver establishment. Sensitivity analysis under extreme conditions
(+25% rainfall, +50% immaturity factor) confirms FoS = 1.23 > 1.2, proving robust resilience. The framework offers three distinct contributions: the first mechanistic “dripline bio-shield” concept using strategically placed Vetiver; the first quantitative linkage between monsoonal soil erosion and pile foundation embedment in agrivoltaics; and a transparent,
software-free methodology calibrated for Bangladesh’s hill slopes. Practical recommendations are provided for engineers, policymakers, and regulators deploying solar infrastructure in monsoon-affected regions.

Sustainability in Higher Education Institutions (HEIs): Two-Step Assessment Approach for Least Developed Countries

Globally recognized sustainability assessment tools (SATs) for higher education institutions (HEIs) predominantly focus on performance outcomes rather than institutional readiness for sustainability transformation, thereby overlooking the unique challenges faced by HEIs in least developed countries (LDCs). This conceptual paper proposes theory based conceptual framework for assessing HEI readiness for SDG-oriented transformation in resource-constrained settings. The study conceptualizes sustainability readiness as an institution’s ability to identify, absorb, implement, and maintain SDG-oriented transformation, drawing on Institutional Theory, Resource-Based View and Organizational Readiness Theory. The paper introduces a novel Two-Step Assessment Approach for sustainability in HEIs, first assessing institutional readiness, followed by performance evaluation, specifically designed for LDC contexts. The proposed framework comprises of six interconnected dimensions: government preparedness, financial readiness, infrastructure readiness, academic and research readiness, socio-cultural readiness, and monitoring and reporting readiness. The framework contributes to advancing sustainability literature from outcome-based assessment toward readiness-based institutional transformation in HEIs across LDC contexts, particularly in South Asia and Sub-Saharan Africa. The paper further proposes theoretical propositions to guide future empirical validation and policy formulation.

Bibliometric Analysis and Knowledge Discovery Patterns in Cybersecurity-Based Fraud Detection Systems in the Banking Sector

As financial crime becomes increasingly digital, financial institutions are increasingly relying on cyber-security-enhanced fraud detection systems to combat increasingly sophisticated digital financial crime. Over the last decade, there has been a growing body of scholarly work in this field, and bibliometric analysis has become useful to map the intellectual structure, thematic development, and knowledge discovery process in this field. This paper reviews the current literature on cyber-security based fraud detection in banking using a comprehensive bibliometric and systematic review with the aim of synthesizing the results of fourteen peer-reviewed and indexed studies published mostly in the period 2023–2026. The review brings together the evidence on trends in publications, prevalent research areas (such as AI, machine learning, blockchain, regulatory compliance), citation, co-authorship networks and geographical concentration of research. It also investigates the algorithmic landscape reported over systematic reviews covering traditional machine learning, deep learning, ensemble and hybrid detection algorithms to identify common knowledge discovery patterns that appear from the co-occurrence of keywords and thematic clustering. The last section of the survey highlights research gaps such as coverage of financial-institution databases in multiple languages, limited research on digital-forensic and intrusion-detection topics, interpretability problems in the “black-box” models, and the lack of bibliometric studies on specific financial-institution types. For future research, directions for research and practice are presented to help scholars and practitioners in building a more comprehensive knowledge production and learning model on cyber-security enabled fraud detection in bank.

AI-Driven Resume Classification and Job Recommendation System

The paper presents an AI-based system that enhances job-matching by using machine learning (ML) and natural language processing (NLP) techniques to improve resume screening and job recommendations.

A Comparative Evaluation of GPT-4o-mini and GPT-3.5 Turbo for Natural Language Spreadsheet Formula Generation

This paper presents the development and evaluation of a formula generation tool integrated into Google Sheets using GPT-4o-mini, with a comparative analysis of its efficiency and cost against GPT-3.5 Turbo model.

Beyond Point Estimates: Stability-Aware Interpretable Machine Learning for Drug-Induced Autoimmunity Prediction

This paper addresses both gaps directly on a public UCI
dataset, the RDKit-descriptor release underlying InterDIA.
Our contributions are:
1) A leakage-safe, nested cross-validation benchmark of
six algorithm families (linear, bagged-tree, two boosting
variants, and an imbalance-aware ensemble), with
algorithm ranking corroborated by a Friedman/Nemenyi
statistical test, and confirmed on the official heldout
test set.
2) A bootstrapped feature-selection stability analysis using
Jaccard similarity and Nogueira’s chance-corrected
stability index, quantifying how much ”important”
descriptors shift under resampling in this highdimensional,
small-sample regime.
3) A systematic SHAP-vs-LIME cross-validation, comparing
global and per-instance rank agreement and top-k
feature overlap, to test whether aggregate XAI agreement
masks local disagreement.
4) A descriptor-space matched-pair mechanistic sanity
check, inspired by InterDIA’s structurally matched case
studies and by matched molecular pair analysis in
medicinal chemistry, used here as a proxy robustness
probe rather than a validated structural-analogue
analysis.

A Two-Stage Global-Scan Fuzzy Logic MPPT for Grid-Connected PV Systems Under Partial Shading Conditions

Most fuzzy-MPPT papers only handle uniform irradiance, where a single local controller is enough. Under partial shading, that same controller becomes actively misleading — it will converge confidently and with low ripple onto the wrong peak, silently losing 20–30% of available power.

This paper’s contribution is a low-overhead fix for that specific failure mode, positioned between two existing extremes:

Cheaper than metaheuristic GMPPT (PSO/GWO-fuzzy hybrids): no swarm size, inertia weights, or iteration tuning — just a one-time bounded voltage sweep (20 points × 10 ms) triggered on a shading event.
More reliable than plain local MPPT (P&O or fuzzy alone): guarantees the global peak is found rather than whichever peak the local gradient happens to lead to.

Its most concrete, defensible finding is a subtle failure mode most papers gloss over: correctly locating the global peak isn’t sufficient — a purely local fine-tracking stage can still drift back off it across a shallow valley created by bypass-diode kinks. The fix (bounding the local controller’s excursion window to the scan’s own resolution) is a small, cheap, and reusable design rule, not just a result specific to this array.

So the contribution is narrow and practical rather than a fundamentally new algorithm: a documented, minimal recipe for retrofitting global-search capability onto an existing local fuzzy controller, plus the reason a naïve version of that retrofit quietly fails

An Offline, Voice-Enabled Mobile System for Leaf Disease Detection in Low-Resource Agricultural Settings of Bangladesh

The main contribution is a complete offline, voice-first pipeline for crop leaf disease diagnosis, not just a compressed classifier but full Bangla voice guidance built for low-literacy users, running entirely on entry-level Android hardware with no internet dependency. The quantized MobileNetV2 model (3.8 MB, 92.8% accuracy) delivers a full capture-to-spoken-diagnosis cycle in under 400ms, addressing a gap in prior mobile plant-disease systems, which are typically cloud-dependent, text-based or English-only. The system was field-validated with 10 real smallholder farmers of mixed literacy, achieving 100% unassisted task completion and 4.7/5 satisfaction — early evidence that offline, voice-first agricultural AI is both technically and practically viable in rural Bangladesh.

Blockchain-enabled Secure Central Bank Reserve Management: A Framework on a Single Case Study

This research adds to the information systems scholarship by broadening the concept of digital empowerment to the domain of central-bank reserve management. Additionally, it offers central banks and financial institutions a useful design guidance for enhancing their cyber resilience within the constraints of existing international payment systems.

XMF-FL: An Explainable Multimodal Feature Fusion Federated Learning Framework for Multi-Class Kidney Disease Classification

We proposed an explainable multimodal feature fusion FL framework for privacy-aware multi-class kidney disease classification. The framework integrates a fine-tuned transfer learning model (i.e., FT-EfficentNetB0) and a customized CNN model using a multimodal dataset (i.e., kidney CT images and corresponding clinical data) in a heterogeneous environment to enhance the
performance of kidney disease. In addition, the framework explains predictions through XAI methods to provide accurate, trustworthy, and privacy-preserving kidney disease diagnosis.

The key contributions of this study are presented as follows:
1) We propose an explainable multimodal feature fusion federated learning framework called XMF-FL for classifying multi-class kidney diseases into four categories while preserving patient privacy.
2) We implemented a heterogeneous data distribution simulation environment with multiple healthcare clients to ensure the performance of the framework in a real clinical environment.
3) The Explainable AI (XAI) integrates with the framework to identify key imaging and clinical features, enhance model transparency, and ensure the trustworthiness of clinicians.
4) We compared the proposed XMF-FL framework with
the state-of-the-art models to validate the effectiveness.

A state-aware progressive Bézier path modifier for cost-aware 3D UAV navigation

This work proposes a progressive, state-aware downstream layer of Bézier trajectory modification framework for agile UAVs operating in complex three-dimensional environments. The main contributions of this research are:

1) A motion-state-aware Bézier trajectory modification approach that incorporates the UAV’s instantaneous velocity magnitude and heading to initialize Bézier control points and generate locally executable curved trajectories

2) A three-stage progressive control-point search strategy that adaptively explores candidate trajectory geometries, progressively extends the Bézier segment across successive waypoint corners based on collision-free feasibility and reduces unnecessary candidate evaluations through early horizon termination

3) A multi-objective candidate evaluation framework that balances directional continuity, curvature-induced lateral-acceleration demand and path efficiency, while enforcing obstacle clearance as a hard feasibility condition.

A Gender-focused Multi-level Data Analysis and Visualization of Academic Research and Collaboration in Bangladesh

The core contributions of this work:
• Granular and Gender Focused Data Investigation.
• Advanced Collaboration Network Exploration.
• Composite Institutional Ranking and Publication Venue List.
• Interactive and Data-Driven Web Platform.

RareGuard: Effective-Number Reweighting with Selective Prediction for Multiclass Intrusion Detection

An intrusion detector may perform well on an
imbalanced test set and still miss the rarest attacks. We study that
tradeoff using the official UNSW-NB15 training and test files.
RareGuard fits a histogram gradient-boosted classifier with effectivenumber weights chosen on a 20% validation partition, then applies a
confidence threshold fixed on the same partition. Effective weighting
increases test macro F1 from 0.4924 to 0.5144 and balanced accuracy
from 0.5428 to 0.6198. Inverse-frequency weighting reaches 0.6488
balanced accuracy, but macro F1 falls to 0.4960 and calibration
deteriorates sharply. RareGuard also sacrifices overall accuracy, from
0.7649 to 0.7264, while expected calibration error increases from 0.0661
to 0.0872. A validation threshold aimed at 80% coverage accepts 85.77%
of test flows and yields 0.7990 conditional accuracy. RareGuard is
therefore not uniformly better than the alternatives. It is better viewed as
a first-stage policy for settings that prioritize class-balanced recognition
and can route low-confidence flows to another process. The executed
pipeline is documented with checksums, settings, adverse results, perclass recall, model size, and latency.

Harnessing Machine Learning for 6G-Integrated Network Security: Opportunities and Challenges

This research provides a comprehensive analysis of the role of machine learning in enhancing security for future 6G networks. It reviews existing ML-based security approaches, identifies key opportunities such as intelligent intrusion detection, adaptive threat prediction, automated response mechanisms, and privacy-preserving security solutions. Additionally, the study highlights major challenges, including adversarial attacks, data privacy issues, computational complexity, and standardization limitations. The paper proposes future research directions toward developing secure, intelligent, and autonomous 6G network architectures through advanced AI-driven security frameworks.

LiMCor-Net: An Extremely Lightweight Multi-Scale Context Refinement Network for Liver Segmentation from Abdominal CT Images

We built a extreamly lightweight model for liver segmentation

Designing a Prompt Framework for Generative AI–Based Critical Thinking Support in SSC Mathematics Learning

As Generative Artificial Intelligence (Generative AI) gains traction, it is poised to offer
innovative solutions for teaching and learning mathematics, particularly in the form of intelligent tutoring systems. But most current AI tutoring systems are designed to output
answers rather than to help students through a sequence of reasoning and problem-solving
steps. For this reason, their contribution to the development of critical thinking is limited,
a very important goal of mathematics education. In addition, current prompt engineering
techniques tend to be optimized for a single prompt rather than offering a complete pedagogical structure based on validated educational theory.
This study aims to develop a structured prompt framework for a Generative AI tutoring
system for facilitating critical thinking in the learning of mathematics in Secondary School
Certificate (SSC). The framework combines the problem solving approach of Pólya with
prompts representing the four stages of the problem solving process: understand the problem, make a plan, execute the plan, look back. Prompts for learners’ guidance, evaluation of
their responses, adaptive feedback, and advancing to the next stage are included as distinct
sections in each stage to facilitate the learners’ active reason and reflective learning.
The study is guided by Design and Development Research (DDR) methodology consisting of four phases namely problem identification by literature review, design of framework,
design of prototype and evaluation by expert opinion. A prototype of AI mathematics tutor based on the proposed framework is built, and the framework is assessed by experts in
computer science and mathematics education through questionnaires and semi-structured
interviews. The assessment is on the clarity, pedagogical relevance, usability and potential
of the framework to support critical thinking.
This inquiry offers a pedagogically informed prompt model incorporating the principles of mathematics learning and of prompt engineering for Generative AI. The proposed
framework offers concrete design considerations for designing AI tutoring systems to support structured mathematical thinking and critical thinking, especially in the context of SSC
mathematics learning.

3-Phase Full-Wave Rectifier: A Review on Topology, Development, Applications, and Future Directions

This is a review paper. I highlighted the development process of a three-phase full-wave rectifier in electronics.

Light Intensity Control Using Single-Phase Controlled Rectifier (SCR)

I designed and tested an SCR-based light intensity control system using phase angle control. My work involved evaluating its efficiency, power factor, and harmonic distortion (THD) to demonstrate a practical and affordable solution for lighting control.

An Explainable AI-Based Framework for Cardiovascular Disease Detection and Risk Prediction Using Machine Learning

Cardiovascular diseases (CVDs) are still one of the
leading causes of death in the world, and traditional screening
methods are often unable to detect early warning signs. The
authors propose a dual-purpose machine learning model that is
able to detect the presence of a heart disease and to predict a
continuous cardiovascular risk score from the same 1,035 patient
dataset. For this work, sixteen classification and regression
algorithms have been tested, some of which had never been
used in the reference study it is based on: ExtraTrees, AdaBoost,
Voting (Hard/Soft) ensembles for classification and ExtraTrees,
VotingRegressor, and Gaussian Process for regression. A moder
ate SMOTE strategy was employed to counteract class imbalance,
and a stratified 70/30 train-test split and 5-fold cross-validation
was used to prevent overfitting. ExtraTrees showed the best
classification performance (97.65% ± 0.95% CV; 98.71% test
accuracy; AUC = 0.998), while Linear Regression showed the best
regression performance (CV R² = 0.9794 ± 0.017; test R² = 0.99).
Interestingly, Voting Soft did not outperform ExtraTrees, which
means that the errors of the tree-based models are correlated.
SHAP and LIME gave interpretability on the global (chest pain
type, number of major vessels (ca), exercise-induced angina) and
local level (thalassemia status). The results demonstrate that
a well validated and modestly-enriched structure can achieve
comparable and more accurate performance to previously re
ported benchmark, while being transparent and overfitting proof
resistant.

Machine Learning-Assisted Confinement Loss Prediction for a Highly Sensitive PCF-SPR Sensor

A photonic crystal fiber (PCF) based surface plas-
mon resonance (SPR) sensor is proposed for the simultaneous
detection of two analytes over a wide refractive index (RI)
range. Surface plasmons are excited by a single gold (Au) film
deposited on two open channels formed on the outer surface
of the fiber. Across the RI range
of 1.25– 1.43 RIU, a maximum wavelength sensitivity of 23,200
nm/RIU, a peak amplitude sensitivity of 2235.71 RIU−1 and
a best sensor resolution of 4.31 × 10−6 RIU are achieved,
and the two channels are shown to respond selectively to their
own analytes without noticeable crosstalk. The suggested sensor
with a wide sensing range, excellent detection capability, and
compact structure has great potential for biochemical sensing,
chemical analysis, environmental monitoring, and point-of-care
diagnostics. Furthermore, six machine learning regression models
are implemented to predict the confinement loss, among which
Random Forest exhibits the best predictive accuracy with an R2
of 0.9759.

Water Pixel Dilution Bias: A Systematic Measurement Error in Riverine Spatial Feature Engineering

This study identifies and formalizes Water Pixel Dilution Bias (WPDB), an overlooked measurement error where river water pixels distort surrounding land use calculations in standard circular GIS buffers. We derive an exact mathematical formula to predict this bias and validate it across 11 monitoring stations on two rivers in Dhaka, Bangladesh. We also provide a practical flow aware normalization method to correct this distortion in everyday GIS and environmental data science workflows.

Smart Firefighting Robot with Thermal Imaging and IoT-Based Human Detection for Enhanced Rescue Operations

Fire is an event that can cause a lot of loss of life
and property in residential, industrial and commercial
buildings. Abstract It is important to quickly suppress fires and
identify trapped persons to limit loss of life and property. In this
paper, the design and development of an autonomous robot with
Internet of Things (IoT) technology that can detect the presence
of humans in a fire zone and feed the real time alarm
information to a monitoring server is presented. The proposed
system consists of a master controller – Raspberry Pi and a slave
controller – Arduino, which synchronize the sensing, navigation,
communication, and the firefighting processes. Environmental
awareness and obstacle detection is achieved with the use of gas,
ultrasonic, microphone and camera modules, and the thermal
sensor is used to detect the presence of human. The robot has a
Dry Chemical Powder (DCP) fire suppression capability to
effectively put out fires. Through LoRa SX1278 communication
technology, the long-distance transmission at low power allows
for continuous monitoring of critical data even in difficult
conditions, to a local server. Experimental results prove the
capability of the system to accurately detect human presence,
real sense temperature change and prompt emergency
notification to control center. The proposed solution will provide
a cost-effective, reliable and intelligent solution to firefighting
and rescue missions, minimize human firefighters’ exposure to
dangerous environments and enhance the efficiency of
emergency response.

An ESP32-Based Automated Obstacle-Aware Level Crossing System Using Infrared Sensors

The rail and road traffic interact at important
points at rail level crossings, these are the points where manual
operation can be carried out. The absence of real-time monitoring
usually results in severe accidents, particularly in developing
nations. The design is depicted in this paper. and installation of
an ESP32-powered automated obstacle-aware rural level crossing
system through the use of infrared sensors. The proposed system
identifies that a train is approaching and sends a servo motor to
automatically open and close the gate. and at the same time giving
visual and audible warnings. with LED indicators and a buzzer.
The ESP32 microcontroller monitors sensor inputs in real time
and acts in time. of the gate by human intervention. Experimental
results prove good working, rapid response, and fluency. gate
control in controlled conditions. It is created with low-cost and
easily accessible components, so the system is developed cheaply.
it appropriate to be deployed at an unmanned or semi-manned
railway. crossings. The solution suggested leads to less human
dependency, reduces the reaction time, and improves the situation
with the safety of the entire railway intersection.

A Smart Umbrella Design for Enhancing Thermal Comfort in Outdoor Environments

Extreme heat poses a significant threat to human
comfort and health, particularly for individuals who stay
outside for long durations. In this article, the design and
development of a smart umbrella system are proposed, aiming
to enhance personal thermal comfort in hot environments. The
proposed umbrella integrates an active cooling system powered
by a compact-scale control unit. A pelteir module is used here to
convert heat energy into lower and comfortable temperature,
offering a portable and wearable solution to mitigate heat stress.
The system intelligently manages airflow and temperature
within the shaded zone, thereby increasing overall comfort and
usability during hot conditions. The design prioritizes energy
efficiency, usability, and ergonomic factors to ensure versatility
in urban and rural outdoor settings. Field testing validates the
effectiveness of the system for reducing perceived heat,
demonstrating its readiness for use among pedestrians, outdoor
workers, and vulnerable populations. This paper proposes a
sustainable and scalable approach to climate control at a
personal level through wearable smart infrastructure.

Renewable-Powered EV Charging Stations: A Structured Review of Global Technologies, Challenges, and a Sustainable Deployment Roadmap for Bangladesh

This study provides a structured review of renewable-powered EV charging stations, identifies key research gaps, and proposes a region-specific deployment roadmap for Bangladesh with practical recommendations for sustainable EV charging infrastructure.

Multimodal Deep Learning Approach for Multi-Class Oral Disease Detection and Classification

1.Proposed a Novel hybrid model for multi-class oral disease classification.
2. Combined image features (ViT) with medical contextual knowledge (MedBERT) for improved diagnosis.
3. Applied data preprocessing and augmentation to enhance generalization and reduce overfitting.
4. Achieved 99.71% accuracy, outperforming existing deep learning models.
5. Evaluated performance using precision, recall, F1-score, confusion matrix, and comparative analysis.
6. Demonstrated the model’s clinical potential for accurate and efficient oral disease diagnosis.

ESP32-Based Smart Fire and Gas Hazard Detection with Multi-Channel Alert Mechanisms for High-Risk Urban Buildings

Today, the fire leakage and gas leakage risks are
presenting a life and property threat, particularly in urban areas, where a lot of people are present. Traditional systems are slow to detect hazards, prone to false alarms and are not able to function when there’s a power or internet outage. These issues are addressed by our system. Our solution is based on
the ESP32 fire and gas detection system. To ensure accurate detection, multiple sensors are used in the system, such as the MQ-2 sensor, flame sensor, and DHT11 sensor which is used to measure temperature and humidity. These sensors feed into the ESP32, which analyzes the data and signals to look for any
potential hazards. When it is determined that there is a hazard, the system alerts the users using multiple methods. One way, they are alerted with a buzzer and LEDs if they are around. Moreover, a web dashboard and Telegram bot send live updates
and notifications online. If Wi-Fi is down, the system will transmit the alerts using the GSM module via SMS. We have developed a three-layer system that has ensured rapid and reliable delivery of the warning to the user. The result demonstrates the rapid detection, minimal risk of false alarms and robust performance even in challenging environments. The use of multi-sensor fusion for accurate and immediate hazard identification is one of the
important contributions of this research in the field of fire hazard detection, which will provide a timely alert with online and offline communication channels to the user.

Policy-Driven Deep Learning Framework for Facial Age Verification in Senior Citizen Welfare Programs

One of the main requirements is the eligibility on the basis of age. In most of the elder citizen welfare programs.However, traditional document-based checking systems are still susceptible to manipulation and falsification of claims. This study proposes a deep deterrence age verification paradigm to aid in the equitable distribution of elderly benefits, which is resistant to
fraud. Instead of pre-emptive age discrimination, the problem is subjected to a policy-reformulation. Discrimination is an aligned
binary classification task that differentiates individuals. under 60 years compared to 60 years and above. A large-scale database with over 160,000 images obtained. Two open-source repositories are used in model training. and evaluation.Three convolutional neural network architectures. Comparatively three architectuere,
VGG16, EfficientNet-B3, and ResNet-50, are comparable. that were processed under the same experimental conditions. To address lopsidedness and minimize the chances of abandoning
deserving elderly. persons, a class-conscious training practice of selective. Weighted loss optimization and data augmentation are used. The proposed framework has proven successful after
experimental results. obtains good F1-scores and high accuracy, and at the same time. strong recall among the older age group of 60 +. The findings indicate that the suggested solution offers a policy-congruent, trustworthy, and age verification Automated solution that is a socially responsible one. real-world welfare
systems.

A Tabular Transformer-Based Deep Learning Framework for Flood Prediction Using Long-Term Meteorological Data from Bangladesh

This work suggests a transformer-based SAINT system to predict floods for Bangladesh which outperforms traditional machine learning and deep learning models with an accuracy of 97.83%. It also tackles the issue of class imbalance through the use of the SMOTE method and strengthens the interpretability of the models with the help of explainable artificial intelligence (XAI) using the SHAP approach, which will enable to interpret the contribution of meteorological variables to flood forecasting, giving the model more transparency.

Human Centered Digital Wellbeing through a Multi-Agent Actor–Critic Framework for Social Media Addiction Mitigation

1. Propose a human-centered RL framework to reduce social media addiction while balancing engagement and well-being.
2. Develop a realistic simulation environment modeling user behaviors like mood, fatigue, and break adherence.
3. Introduce a lightweight multi-agent actor–critic model with adaptive intervention strategies.
4. Achieve statistically significant improvements in digital well-being and high validation accuracy (92.6%) compared to baselines.

Fed-OCT-IoT: CBAM-Augmented EfficientNet with Federated Learning and Explainable AI for Retinal OCT Diagnosis

There are three primary obstacles to the automated
classification of retinal diseases using OCT: severe class imbalance, the high expense of transformer-based models, and
complications in gathering patient data from multiple medical
facilities. We address all three obstacles. We suggest utilizing an
EfficientNet-B3 backbone combined with a Convolutional Block
Attention Module (CBAM) for our framework to categorize seven
classes in the OCTDL dataset. To address the imbalance issue, we
employed a focal loss and oversampled certain data. The model
was assessed using a 5-fold cross-validation method at the patient
level along with a held-out test set, and was compared to four
independently trained baseline models. Our model outperformed
the baseline models in terms of accuracy. Component-wise
ablation studies revealed distinct contributions of the spatial and
channel-attention mechanisms. We also examined lighter versions
of federated learning, specifically FedAvg and FedProx for edge
devices in both IID and non-IID contexts. In this scenario,
FedProx did not perform better than FedAvg. We assessed
the model’s effectiveness using the Integrated Gradients and
Occlusion Sensitivity methods. Finally, we propose a theoretical
framework for IoT edge implementation and consider it viable
in terms of efficiency, without the need for physical testing in
person, which can be further investigated in future research.

A Robust Integrity Verification and Recovery Framework for Coarse-to-Fine Text Semantic Communication Against Model Tampering Attacks

This paper proposes the first integrity verification and recovery framework for text semantic communication systems that combines component-level tamper detection, localization, and automatic recovery. Using cryptographic fingerprints and digital signatures, the framework verifies receiver model integrity, detects unauthorized tampering, identifies the affected component, and restores it from a trusted backup. Experiments under AWGN and Rayleigh fading channels demonstrate 100% detection and localization accuracy, with recovery completed in about 96 ms while maintaining reliable semantic communication performance.

Cross-Dataset Generalization Analysis of Transfer Learning Models for Osteoporosis Classification Using Knee X-ray Images

Deep learning techniques have shown promising results for osteoporosis classification using radiograph images. However, the real-world applicability is overestimated by most of the existing studies as they evaluated their model only on a single dataset. To address this limitation, this study investigates the cross-dataset generalization performance of transfer learning models, ranging from lightweight to heavyweight CNN architectures, for knee osteoporosis classification using X-ray images.The study provides insights into the generalization capability of different types of deep learning architectures and may help researchers select suitable models for future osteoporosis classification and medical image analysis tasks.

Enforcing Regulated Fares at the Vehicle Level: An Offline ESP32 Meter for Auto-Rickshaws

Auto-rickshaws carry a large share of short-distance
urban trips in Bangladesh, yet most of them run without any fare
meter. Fares are settled verbally before each trip, and the gap
between what a driver asks and what a passenger expects to pay
is a common source of arguments. Commercial GPS taximeters
do not fit this market: they cost too much, draw too much
power, and depend on cellular data that many drivers cannot
afford. This paper presents a low-cost fare meter built around
the ESP32-WROOM microcontroller that runs fully offline. The
system measures distance from wheel rotation using an FC-03
optocoupler sensor read through a hardware interrupt, converts
pulse counts to distance with a calibration constant, and computes
the fare from a rate structure based on the government-notified
auto-rickshaw tariff. An L298N motor driver moves the prototype
chassis, an LM016L LCD shows the running distance and fare
to both driver and passenger, and a battery pack with a TP4056
charger and an LM2596S buck converter powers the unit. A
three-state controller (IDLE, ACTIVE, TRIP END) manages the
trip cycle and returns the meter to its starting state after each
trip. The fare model implemented in firmware follows a 20 tk
minimum with a 12 tk/km rate and rounds the display to the
nearest 5 tk, which matches how fares are actually settled on the
street. The prototype demonstrated stable pulse counting, correct
motor control, consistent power delivery, and a fare display that
tracked the measured distance during operation.

Archives: Paper Submission