Efficient classification of ventricular arrhythmias using feature selection and C4.5 classifier

The occurrence of sudden cardiac arrest (SCA) leads to a massive death across the world. Hence the early prediction of ventricular tachycardia (VT) and ventricular fibrillation (VF) becomes vital to prevent from ventricular arrhythmia. In this study, we present a process to detect and classify VT and VF arrhythmias using temporal, spectral, and statistical features. The CU Ventricular Tachyarrhythmia Database (CUDB) and MIT-BIH Malignant Ventricular Ectopy Database (VFDB) databases are used from PhysioNet database for evaluation and comparison of the proposed algorithm. Thirteen time-frequency based features were extracted for a window length of 5 s with an appropriate thresholding to make a feature set. The gain ratio attribute evaluation has been used for potential utilization of the informative features by ranking them according to their individual evaluation weightage. Classification of selected features for VF, VT, and normal sinus rhythm (NSR) is done by using cubic support vector machine (SVM) and the C4.5 classifiers. Assessment of this process is done on 57 records of electrocardiogram (ECG) signals and the result shows that the proposed method achieved a sensitivity of 90.97%, specificity of 97.86% and accuracy of 97.02% in C4.5 classifier, which was better than the obtained results of cubic SVM having an accuracy of 92.23%. This study demonstrated that by using informative features and classifying them with C4.5 algorithms, the system data could be an aid to the clinician for precise detection of ventricular arrhythmias.