ECG Heartbeat Classification for Cardiac Disease Detection using 1D-CNN with Explainable AI

This paper presents a lightweight (814,215 parameters) 1D-CNN for five-class ECG heartbeat classification on MIT-BIH Arrhythmia dataset with 99% test accuracy and macro-F1 of 0.9195. SMOTE tackles extreme class imbalance without polluting test data, while gradient-based Saliency Maps and handcrafted clinical features (R-peak, QRS duration) offer Explainable AI interpretability a combination not provided by similar previous work on this dataset.