Atrial Fibrillation Classification: A Hybrid Approach Integrating Support Vector Machines and Feature Selection for Reduced Computational Complexity

Atrial Fibrillation (AFib) is an important cardiac arrhythmia marked by an irregular and frequently accelerated heartbeat which increases the risk of stroke and other cardiovascular complications. Detecting AFib promptly allows for timely intervention along with the treatments like Cardioversion therapy, medications and Surgery or catheter procedures might help mitigate these risks. Wearable devices and mobile health applications make use of single-lead ECG, allowing for immediate identification and response but most of the devices are based on high computation algorithms leading to a rise in the overall cost of devices. The proposed methodology automatically classifies ECG signals into AFib and Normal using Machine Learning algorithm. To increase the diversity in the dataset, this study has considered 5822 ECG signals from two prominent databases, steps like feature engineering based on Variable Clustering were used to handle multicollinearity. Different Machine Learning (ML) models were assessed, encompassing Random Forest, Logistic Regression, SVM, and Adaboost. Among these, SVM emerged as the preferred choice due to its outstanding performance, achieving a F1 score of 0.95. The reliability of this evaluation was confirmed through a 10-fold cross validation method and hypertuning the parameters using the Grid Search method. This study can bridge the gap in Atrial Fibrillation diagnosis and potentially revolutionize healthcare by offering trustworthy and accessible diagnostic procedures.