Evaluation of Extreme Learning Machines for Detecting Heart Diseases

Abstract

Currently, cardiovascular diseases are the leading cause of human death according to the World Health Organization. Their prediction allows doctors to indicate preventive measures to their patients and perform procedures on time. In this research, the performance of different Extreme Learning Machine (ELM)-based algorithms applied to the binary classification problem of the heart's state (healthy or sick) was evaluated. The following ELMs were used: the basic model, regularized, weighted, and multi-layer. The experiments were carried out in a MATLAB programming environment and a mid-range laptop. To evaluate the models' performance, the accuracy (Acc), the geometric mean (G-mean), and the execution time of the algorithms were used, comparing the results with other classifiers reported in the literature. In this research, it is proposed to use a Weighted ELM (W1-ELM) due to its acceptable accuracy of 0.81 and its low training complexity compared to deeper models such as Convolutional Neural Networks.