PERFORMANCE EVALUATION OF ARTIFICIAL NEURAL NETWORKS FOR CARDIAC ARRHYTHMIA CLASSIFICATION

In this paper an effective and most reliable method for appropriate classification of cardiac arrhythmia using automatic Artificial Neural Network (ANN) has been proposed. The results are encouraging and are found to have produced a very confident and efficient arrhythmia classification, which is easily applicable in diagnostic decision support system. The authors have employed 3 neural network classifiers to classify three types of beats of ECG signal, namely Normal (N), and two abnormal beats Right Bundle Branch Block (RBBB) and Premature Ventricular Contraction (PVC). The classifiers used in this paper are K-Nearest Neighbor (KNN), Naive Bayes Classifier (NBC) and Multi-Class Support Vector Machine (MSVM). The performance of the classifiers is evaluated using 5 parametric measures namely Sensitivity (Se), Specificity (Sp), Precision (Pr), Bit Error Rate (BER) and Accuracy (A). Hence MSVM classifier using Crammers method is very effective for proper ECG beat classification.