Amar Surokha: An Integrated Mobile Application for Spatiotemporal Crime Prediction and Emergency Contact Management

Personal safety and emergency medical response
remain critical challenges in densely populated developing-country
such as Bangladesh, where rapid urbanization has been accompa-
nied by rising crime rates and delayed emergency assistance. This
paper presents Amar Surokha, an integrated mobile application
that unifies location-based crime prediction, proximity-aware
emergency response, and blood bank management within a
single platform. An XGBoost classifier was trained on 5,012
preprocessed records drawn from the Bangladesh Crime Dataset
(CrimeDataBD), using location, temporal, and weather-derived
features to predict one of three crime categories: Violent Crime,
Body Found, and Murder. XGBoost was benchmarked against a
Random Forest baseline under an identical pipeline; XGBoost
achieved the best overall performance, with 52.44% test accuracy
and a macro-averaged F1-score of 0.5192, validated through
5-fold stratified cross-validation (mean accuracy 51.43% ±
2.58%). The system was implemented using Flutter for cross-
platform mobile delivery, Firebase for real-time data storage,
the OpenRouteService API for road-network-based proximity
ranking of emergency contacts and blood donors, and a Flask
REST API deployed on Render for cloud-hosted model inference.
Results indicate that the proposed system is technically feasible
and provides a reproducible reference architecture for integrating
machine learning with mobile emergency-response infrastructure
in resource-constrained environments.