Citation

Mohajan, Aritra and Fahim, Abu Monsur Mohammad and Tushe, Ummay Ayman and Rabbi, Riadul Islam and Tusher, Ekramul Haque and Liew, Tze Hui (2025) Hybrid CNN-XGBoost Framework for Interpretable ECG Arrhythmia Classification with SHAP-based Analysis. In: 2025 8th International Conference on New Media Studies (CONMEDIA), 14-17 October 2025, Malacca, Malaysia.

Text 16.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Accurate and interpretable detection of cardiac arrhythmias from electrocardiogram (ECG) signals is pivotal for effective cardiovascular disease management. We propose a novel hybrid architecture combining multi-layer convolutional neural networks (CNNs) with eXtreme Gradient Boosting (XGBoost) for automated ECG arrhythmia classification. Our method harnesses CNNs for hierarchical feature extraction and leverages XGBoost classifiers alongside SHapley Additive exPlanations (SHAP) for enhanced interpretability. Evaluated on the comprehensive MIT-BIH Arrhythmia Database with stratified 4-fold cross-validation, our framework achieves a mean accuracy of 98.9%, outperforming pure CNN and XGBoost models. SHAP and gradient-based saliency analyses elucidate lead-wise and temporal feature importances, aligning with clinical domain knowledge and facilitating transparent model decisions. These results underscore the applicability of hybrid, explainable models in advancing automated, trustworthy ECG analysis.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	ECG arrhythmia classification, hybrid deep learning, CNN, XGBoost, SHAP, interpretability, MIT-BIH database
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD28-70 Management. Industrial Management
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	20 Apr 2026 03:04
Last Modified:	20 Apr 2026 03:04
URII:	http://shdl.mmu.edu.my/id/eprint/15762

Downloads

Edit (login required)