Enhancing Arrhythmia Classification Performance using Hybrid CNN and SVM

Abstract

Cardiac arrhythmia, a disorder in the heart's rhythm, can lead to serious complications. This study presents a hybrid cardiac arrhythmia classification model that integrates a one-dimensional Convolutional Neural Network (1D-CNN) with a Support Vector Machine (SVM) classifier to improve the recognition of ECG heartbeat patterns. The model was evaluated using the MIT-BIH Arrhythmia dataset available on Kaggle. Experimental results show that the hybrid 1D-CNN-SVM architecture achieves 96.84% accuracy, outperforming the baseline 1D-CNN with SoftMax, which attained 83.64% accuracy. The hybrid approach also demonstrates substantial improvements in class-balanced metrics, with Macro Precision increasing from 0.58 to 0.81 and Macro F1-Score rising from 0.63 to 0.85. These results indicate that the proposed architecture not only enhances overall predictive performance but also delivers more stable and reliable classification across all arrhythmia categories, particularly minority classes prone to misclassification. By effectively reducing false positives and maintaining a stronger precision–recall equilibrium, the model offers improved clinical relevance for automated ECG analysis. Future research may further optimize CNN–SVM hyperparameters, validate generalization across diverse ECG datasets, and explore deployment on low-power wearable monitoring systems.