An Evaluation of Machine Learning Classifiers for Detection of Myocardial Infarction Using Wavelet Entropy and Eigenspace Features

This study aims to compare the utility of several machine learning (ML) algorithms for predicting Myocardial infarction (MI) from multi-lead electrocardiogram (ECG). Wavelet-based features are used for the evaluation, as wavelet decomposition of ECG segregates clinical components into different subbands. The pathological changes are reflected in these sub-bands and are captured by wavelet entropy and eigenspace based features. The Physikalisch-Technische Bundesanstalt (PTB) diagnostic database consisting of healthy and different types of MI is used for evaluation. The results show that the K-nearest neighbour (KNN) algorithm with the nearest neighbour set to six obtained the best result among the ML classifiers with F1-score of 0.97 and 0.94 for MI detection and localization respectively. The performance of the support vector machine (SVM) with radial basis function (RBF) kernel obtained an F1-score of 0.96 and 0.92 for detection and localization respectively. The results obtained using classical ML classifiers were also compared with the performance of 1-dimensional convolutional neural network (CNN).