Performance Analysis of Ensemble and DNN Models for Decoding Mental Stress Utilizing ECG-Based Wearable Data Fusion

The growing number of stress-related illnesses highlights the critical need for accurate and efficient stress detection techniques. Stress detection using wearable devices has gained significant attention due to its potential for promoting well- being. With the increased use of smartwatches, there is a scope to utilize this wearable for stress detection and management. Nowadays, the advancement of machine learning and deep learning techniques has significantly simplified the tasks of detection and classification. This study conducts a performance analysis of Ensemble and Deep Neural Network (DNN) models for real time stress detection leveraging a fusion of Electrocardiogram (ECG) data, a well-known indicator of stress-induced physiological changes, sourced from two different datasets of smartwatch data. The models undergo training and testing using four different combinations of both datasets, and the optimum results are subsequently compared. The study demonstrated an accuracy of 94.66%, coupled with a Matthews Correlation Coefficient (MCC) score of 86.01 %, when utilizing a single dataset. Conversely, employing a data-level fusion approach with two datasets yielded an accuracy of 91.93 % and an MCC score of 79.3%. The results highlight a promising method for stress detection using real-time data obtained from smartwatches.