Human Stress Classification Using Electroencephalogram (EEG) in Response to Standup Comedy Clips

Abstract

Nearly everyone encounters stress at some point in their life. An individual’s stress load can be estimated using a valid and reliable stress assessment. In this study, stand up comedy clips in native and non-native languages are used as a stimulus to study the reduction in stress levels. The electroencephalogram (EEG), which reflects brain activity and is frequently employed in clinical diagnosis and biomedical research, serves as the main signal. An EEG dataset is generated from thirty participants using a single channel Neurosky Mindwave 2 mobile headset. The electrical activity of the brain is captured as the participants watch various comedy clips. A state and trait anxiety questionnaire is used to obtain a subjective measure of stress level of participants. The single-channel EEG data being an extremely noisy, non-stationary, and a non linear signal is filtered using the Savitzky-Golay filter. Ten features from the wavelet, time-frequency, and time domains were used to classify stress using each domain. Long Short Term Memory (LSTM), Random Forest, eXtreme Gradient Boosting (XGB), and ExtraTree classifiers were used where the highest accuracy achieved was 84.29% with the ExtraTree classifier. Our findings indicate that only two classes (stressed, and Non-stressed) can be classified for a single-channel EEG device. Where non-native and native language comedy clips have obtained the maximum individual accuracy of 84.29% and 78.32%. It is evident from the results that English comedy has more influence on stress level reduction as compared to Urdu comedy.