Computational Models of Stress in Reading Using Physiological and Physical Sensor Data

Abstract

Stress is a major problem facing our world today and it is important to develop an objective understanding of how average individuals respond to stress in a typicalactivity like reading. The aim for this paper is to determine whether stress patterns can be recognized using individual-independent computational models from sensor based stress response signals induced by reading text with stressful content. The response signals were obtained by sensors that sourced various physiological and physical signals, from which hundreds of features were derived. The paper proposes feature selection methods to deal with redundant and irrelevant features and improve the performance of classifications obtained from models based on artificial neural networks (ANNs) and support vector machines (SVMs). A genetic algorithm (GA) and a novel method based on pseudo-independence of features are proposed as feature selection methods for the classifiers. Classification performances for the proposed classifiers are compared. The performance of the individual-independent classifiers improved when the feature selection methods were used. The GA-SVM hybrid produced the best results with a stress recognition rate of 98%.