Decision Support System for Predicting Ventricular Arrhythmias Using Non-linear Features of ECG Signals

Abstract

Automated methods using computer-aided decision-making process are effectively used for timely detection of VAs which are the most life-threatening conditions. In this work, we have proposed a decision support system for detection of ventricular arrhythmias (VAs) with a low computational complexity using hybrid features derived from three different transform techniques i.e. DWT, EEMD, and VMD. The methods mainly consist of a windowing technique, signal decomposition, feature extraction, and classification. About 24 time–frequency based features were extracted with 22,721 × 24 and ranked for selection of higher ranked features having maximum information of disease. The reduced feature set consists of only 6 number of highly ranked features which are then classified with SVM and decision tree classifier for efficient recognition of VAs. Aim of the reduction in data size is to reduce the computational time. Our proposed method achieves high classification accuracy in hybrid-based features and by reducing the feature dimension, it reduces the computational complexity significantly. The accuracy of 99.83% and computational time of 4.82 s is achieved when considering all 24 features. In reduced feature set, an accuracy of 99.62% with a very less computational time of 2.71 s was obtained for decision tree classifier which indicates the importance of selecting the important features for classification of VAS. With superior classification accuracy and low computation complexity, this system can be utilized in clinical practice for the recognition of ventricular arrhythmias.