A New Physiology-based Objective Mental Stress Detection Technique with Reduced Feature Set and Class Imbalanced Dataset Management

Abstract

With the rapid changing world of work habits, social interaction, and ambitious lifestyles of society, mental stress detection assumes much importance in maintaining good quality of life. In this respect, continuous monitoring of mental stress using physiological signals can be a good solution. In this work, a new method for mental stress detection is described using two common physiological signals, viz., electrocardiogram (ECG) and galvanic skin response (GSR). The statistical feature extraction from a 10 s segment was performed by wavelet packet decomposition, followed by selection of optimum features by extremely randomized tree (ERT). For uniform training of the classifier, the dataset imbalance is adjusted with adaptive synthetic minority oversampling (ADASYN) technique followed by a multi-class random forest (RF) classifier. The proposed technique was evaluated with the WeSAD dataset to detect three-levels of mental state, viz., baseline, stress and fun. The system achieved an overall accuracy, F1-score, precision and recall of 97.08%, 0.98, 0.98 and 0.99 respectively. The results outperform published work in respect of number of sensors (signals), feature set, accuracy, and segment length for detection. The proposed technique may be extended to develop a wearable healthcare system.