Enhancing Clinical Psychology Practice through Data-driven Machine Learning Monitoring Systems

Mental health is a critical global challenge, with significant societal impacts. Machinelearning has emerged as a valuable and innovative approach to enhancing psychologicalassessment and intervention. This study focuses on leveraging mental health indicators fordepression and anxiety to develop and test a data-driven alert system for a workplace-centredpsychological monitoring system. Synthetic cases data (N=192) were generated throughclinically-based criteria, while real evaluation data (N=489) were derived from evaluationsprovided by experts (N=46) across five mental health diagnostic fields. We adopt amulti-layer supervised machine learning approach (MLP) to address various predictive tasks,incorporating SHAP analysis (explainable AI) and Monte Carlo Dropout (uncertaintyestimation) to enhance interpretability and reliability. The study demonstrated that MLPsachieved a commendable performance, with an alert system score of 0.68 on a scale of 0 to 1,and that agreement between evaluators is crucial to train supervised models. We showed thatagreement between experts, although not strong, is higher than that of our MLP models andexperts for medical discharge. These findings highlight the potential of these methods toimprove mental health monitoring in workplace settings, while also underscoring the need formore extensive data collection in real-world environments to further validate and refine thesystem in future applications of Metrikamind’s service.