Multimodal Mental Workload Classification Using Maus Dataset

This study aims to use machine learning to detect stress in individuals to enhance their quality of life, through stress management techniques. We propose supervised learning-based classifiers model for the detection of stress in humans using a MAUS dataset which is multimodal and is publicly accessible. MAUS focuses on the investigation of wearable device-based mental workload (MW) assessment system. The database contains records of fingertip PPG, galvanic skin reaction (GSR), electrocardiography (ECG) and wrist-worn PPG. For multimodal fusion, the normalized features of ECG and PPG (fingertip and wrist) are concatenated, and early fusion based classifier methods are applied. The accuracy and F1 score for combined modalities namely, ECG, PPG (fingertip) and PPG (wrist) is reported at 75.7% and 73.9% using AdaBoost Classifier.