Predicting Depression and Anxiety Using Digital Phenotyping and Machine Learning: Toward Personalized Mental Health Monitoring

Background and Aims: Depression and anxiety are among the most prevalent mental health disorders globally, yet their diagnosis often relies on subjective clinical assessments. Advances in digital phenotyping—collecting behavioral and physiological data from smartphones and wearable devices—offer an opportunity for continuous and objective monitoring of mental health. This study investigates the use of machine learning algorithms to predict levels of depression and anxiety based on passive data collected from digital devices, aiming to develop a personalized and scalable mental health monitoring tool. Methods: A cohort of ۲۵۰ participants installed a mobile app that collected passive data over ۶۰ days, including GPS patterns, screen time, call/text logs, and sleep/activity metrics from wearables. Participants completed weekly PHQ-۹ and GAD-۷ questionnaires. Several models, including random forests, support vector machines, and deep neural networks, were trained to classify depression and anxiety severity levels. Results: The deep neural network model achieved the highest accuracy (F۱-score = ۰.۸۸ for depression, ۰.۸۴ for anxiety). Feature analysis showed that late-night screen time, decreased physical activity, and social withdrawal (e.g., fewer calls/messages) were strong predictors of poor mental health. The system provides weekly risk alerts and recommendations for follow-up care. Conclusion: Digital phenotyping combined with AI enables proactive and personalized mental health support. Our findings support the use of passive behavioral data in augmenting traditional psychiatric assessment and facilitating timely interventions.

کلیدواژه ها:

Digital Phenotyping ، Mental Health ، Machine Learning