ECG arrhythmia classification using R-peak based segmentation, binary particle swarm optimization and absolute euclidean classifier

Abstract

This paper proposes a novel technique to classify arrhythmias from ECG signals using time domain and frequency domain approaches. The ECG signal is pre-processed using Fast Fourier Transform (FFT). It is then segmented into beats after detecting the R-peaks. The Discrete Cosine Transform (DCT) and Discrete Wavelet Transform (DWT) used for Feature Extraction pack most information in fewest coefficients. The Binary Particle Swarm Optimization (BPSO) algorithm used for Feature selection reduces dimensionality by selecting subset of original variables. The proposed Absolute Euclidean Classifier (AEC), which uses the absolute values of the features instead of their actual values, is found to improve the Classification Rate significantly. Feature Extraction using DCT/DWT and Feature Selection using BPSO, together with pre-segmentation process results in an improved Classification Rate and a reduced number of selected features for the proposed Arrhythmia Classification system. Experiments conducted on MIT-BIH Database show an enhanced performance as compared to other systems. © 2013 Springer.