Comparison of Machine Learning Approaches for Physiological States Classification Using Heart Rate and Pulse Rate Variability Indices

Abstract

In this work, we investigate the feasibility of classifying physiological states including conditions of postural and mental stress using heart rate variability (HRV) and pulse rate variability (PRV) time-, frequency- and information-domain indices. The performance of four different machine learning algorithms, i.e. Linear Discriminant Analysis (LDA), Support Vector Machines, Neural Networks (NN) and k-Nearest Neighbors, were compared, with and without prior applying the minimum Redundancy Maximum Relevance (mRMR) algorithm for feature selection. Analyses were conducted on 76 young healthy subjects under three different conditions (supine rest, orthostatic and mental stress). Results evidence higher accuracy for HRV indices if compared to PRV and better classification performance of orthostatic stress. The highest accuracy has been achieved by the NN algorithm on HRV time series, with a value of 90.9% after feature selection, while LDA is the best algorithm for PRV features (81.8%).