Comparison of machine learning algorithms to predict psychological wellness indices for ubiquitous healthcare system design

Abstract

For ubiquitous healthcare service delivery, psychological wellness indices have been developed. A psychological wellness index integrates the survey results that measure stress, depression, anger, and fatigue. The current model is based on a multiple regression method and manually constructs a cause and effect model of the psychological wellness. However, this constructed model depends upon the survey responses. The relationship between these survey responses and psychological wellness indices are not linear due to data imbalance. When any data inconsistency exists, the reliability of the model decreases and eventually cost of maintenance on model revision increases. Also, when new variables or data entries are considered, the entire model should be constructed again. This paper examines the feasibility of machine learning algorithms to predict the psychological wellness indices based on the reconstructed responses. In this paper, four machine learning algorithms including multi-layer perceptron, support vector regression, generalized regression neural network, and k nearest neighbor regression, are compared and the experiment results are presented.