Machine Learning based Assessment of Mental Stress using Wearable Sensors

Stress is recognized as a strong factor linked to severe health conditions like hypertension, cardiovascular diseases, and diabetes. With the growing emphasis on wearable health monitoring, numerous investigations have been conducted into the feasibility of leveraging diverse physiological markers to detect stress. This research endeavors to conduct classification using physiological data, drawing from the readily accessible WESAD (Wearable Stress and Affect Detection) dataset. The primary goal is to employ this dataset to develop models capable of predicting stress based on physiological indicators. In this paper, a model is designed to enhance the accuracy of stress level detection through the application of the Synthetic Minority Oversampling Technique (SMOTE). The purpose of SMOTE is to rectify the issue of imbalanced datasets by oversampling the minority class. To handle the imbalanced nature of the data, this study adopted the SMOTE technique to effectively balance the dataset groups.