Classification of Mental Stress Using an Explainable Machine Learning Approach

Mental stress, characterized by feelings of strain, tension, or pressure, poses significant implications for individuals’ well-being and overall health. Timely and accurate detection of mental stress is crucial to diagnose and prevent the problem. Despite successful past classifiers developed to predict mental stress, there is also a need to justify these predictions. This work aims to classify mental stress using the machine learning predictive classifiers; to interpret these classifier models using explainable machine learning methods; and to find the driving features of the developed models to assist non-technical medical personnel for early medical diagnosis and timely medical intervention. Experiments have been performed on biometric data obtained from nurses during the COVID-19 pandemic period by using a number of classifiers. The model performances were evaluated in terms of accuracy, precision, recall, and F1 metrics. According to the experiment results, the Random Forest algorithm was better than Logistic Regression, K-Nearest Neighbors, and Decision Tree. Next, explainable model techniques like SHAP and LIME are employed based on a balance between prediction strength and model interpretability to find features that best contribute to stress prediction.