<![CDATA[This systematic literature analysis examines 40 studies (2020–2025) on the use of machine learning to predict mental health using digital activity data. Two research questions are presented: algorithm performance comparison and model effectiveness factor. Data surveys (43,9%) are a more widely used data collection method. Because of its interpretability, Logistic Regression is the most popular (29.3%), whereas Random Forest (26.8%) is best for performance-interpretability. With a rata-rata accuracy of 80.1% ± 4.2% and an AUC of 87.1% ± 1.8%, XGBoost provides superior performance. The best study achieves an AUC >0,98 through feature engineering that canggih using SHAP and recursive feature elimination. Critical success factors include cermat fitur selection, temporal dinamika, cross-validation, and clinical interpretability. Although machine learning has significant potential, there are still challenges with standardization, generalizability, and real-world implementation. Research in the long term requires longitudinal studies, external validation, and standard protocols to realize this technology's potential in improving mental health outcomes.]]>
Copyrights © 2025
