Improving Multimodal Stress Detection Using MAUS Dataset with PPG, ECG and GSR Signal

Stress is the emotional and cognitive strain a person feels when confronted with demanding, challenging, or overpowering events. Stress can be caused by various factors that can have harmful effects on physical and mental health. In this study, we proposed a supervised machine learning-based classifier model for the detection of stress in humans using an MAUS dataset. MAUS is an open-access database that contains records of fingertip PPG, galvanic skin response (GSR), electrocardiography (ECG) and wrist-worn PPG. We extracted different heart rate variability (HRV) features, such as Time domain features and Frequency domain features from ECG, PPG(fingertip) and GSR. Concatenated the features, different classifier methods have been applied. Finally, an accuracy of 89% is achieved by using the Random Forest classifier with Hyperparameter tuning.