EEG Features Selection by Using Tasmanian Devil Optimization Algorithm for Stress Detection

Stress detection is imperative in human–machine interaction systems to monitor the mental health of the person. Various stress detection schemes based on machine learning and deep learning have been presented in the past. However, the performance of the stress detection system is challenging due to inadequate training data, poor feature variability, poor feature selection, and the intricacy of the deep learning framework. This paper provides the stress detection system using multiple electroencephalograph (EEG) signals and a Tasmanian Devil Optimization (TDO)-based feature selection scheme to select important and distinctive features. It uses a lightweight Deep Convolutional Neural Network (DCNN) to improve feature representation and stress detection. The efficiency of the suggested scheme is estimated on the public Dataset for Emotion Analysis using Physiological Signals (DEAP). It provides a stress detection rate of 97.95 and 89.75% with and without feature selection. It is observed that the proposed feature selection scheme provides a significant improvement in the traditional feature selection techniques.