Mental Stress Detection using EEG and Recurrent Deep Learning

Mental stress is a major health problem and affects the individual’s capability to perform in day-to-day life. This paper proposes a novel deep-learning (DL)-based-artificial intelligence (AI)-approach that uses electroencephalogram (EEG) data to build an emotional stress state detection model. The EEG data are first processed to extract time and frequency-domain features, which are then supplied to train DL algorithms and identify emotional stress state. Three different DL methods, namely, one-dimensional convolutional neural network (CONVID), bidirectional long-short term memory network (BiLSTM) and bidirectional gated recurrent unit (BiGRU) networks are considered for constructing stressdetection models. The proposed approach is validated using the benchmark dataset-Database for Emotion Analysis using Physiological Signals (DEAP) available freely in the public domain. The DEAP dataset consists of EEG data of 32 participants recorded by exposing them to 40 one-minute-long expressive music video samples along with the respective emotion-ratings. The results showed that among the different developed DL models, the CONVlD+BiLSTM provided the highest emotion detection accuracy of 88.03 % and outperformed the conventional shallow learning approaches.