Ecg Arrhythmia Signals Classification Using Particle Swarm Optimization-Support Vector Machines Optimized With Independent Component Analysis

Abstract

Abstract Cardiac Arrhythmia is one of the serious disorders which are most commonly found among humans larger in number. This study is based on proposing a novel approach for heart (Cardiac) arrhythmia disease classification. Many Machine learning algorithms are implemented for the cardiac arrhythmia classification from which the ECG signal are extracted from MIT-BIH Database. The main objective of this study is to do the classification of ECG signals to the normal and abnormal (Ventricular Tachycardia) category using PSO-SVM optimized with Independent Component Analysis using Genetic Algorithm. The extraction of ECG signal is done with twenty four features consisting of Normal and Abnormal clinical clusters. ECG Signals under these categories are extracted from MIT-BIH Arrhythmia database which is read in terms of P,Q,R,S and T voltage-time parametric signal. Genetic Algorithm and Particle Swarm Optimization together used to enhance the performance of the Support Vector Machine (SVM) classifier. Initially the SVM Classifier is designed and it is optimized by searching for the best parametric value where the discriminate function is tuned to extract the features under the best subsets and as a result the fitness functions which are classified are identified with better optimization. Additionally the PSO-SVM Classifier is allowed to undergo the adaptive mechanism wherein which the optimization factor is allowed to restrict the boundaries of classification of ECG arrhythmia with maximum accuracy by the implementation of Independent Component Analysis Optimization using Genetic Algorithm. The results are experimentally demonstrated with the comparison of PCA, ICA, PSO-SVM with ICA and G-ICA. Sensitivity, Specificity, False Positive Rate, True Positive Rate and Accuracy are the experimental parameters used for the performance metrics comparison to classify for normal and diabetic clinical condition. The parameters yield better results for PSO-SVM-ICA and G-ICA with respect to the above mentioned metrics. The Classification Accuracy is attained with 96% with best optimization strategies by using these hybrid classifiers.