EEG-BASED MENTAL WORKLOAD AND EMOTION DETECTION USING MACHINE LEARNING AND DEEP LEARNING

Abstract

Mental stress has significant impact on critical thinking, problem solving, behaviour, social interaction, and general intelligence. Electroencephalography (EEG) is a simple method which gives an idea about the potential generated on the surface of the brain which helps in understanding the functionality of the brain. EEG finds its use in various bio-medical and bio-informatics research works where the EEG patterns can be used to predict and analyse a person’s mental state and awareness. In this work, we propose an effective approach to detect mental stress. We address two problems in EEG-studies that have generated considerable interest in recent times: (a) Emotion Recognition, and (b) Mental Stress Detection. We make use of publicly available benchmark EEG datasets to investigate these problems and unravel key information from these studies. From the Emotion Recognition study, we formulate an effective emotion recognition approach and highlight the differences between subject-dependent and subject-independent models. We achieve the best average classification accuracy of 93.91%. The stress detection study on publicly available dataset ’EEG During Mental Arithmetic Tasks’ finds the theta and alpha power of EEG signals most indicative of cognitive workload. The study also highlights the significance of frontal EEG channels in mental stress studies. We achieve the best accuracy of 93.05% for this study. We apply the findings of these studies on the EEG recordings that we gather from students of Delhi Technological University. We obtain the best classification accuracy of 90% for this study. We propose an efficient EEG-based stress detection system that can be used to determine stress in students in real world.