Early Detection of Relapse in Schizophrenia Using Passive Smartphone Sensing: A Longitudinal Digital Phenotyping Study

2"}This preprint presents a longitudinal study investigating the use of passive smartphone sensing and machine learning for early detection of relapse in schizophrenia. Behavioral data including mobility patterns, activity levels, and communication metadata were analyzed using predictive models to identify clinically meaningful changes prior to relapse events. The study demonstrates that deviations in behavioral patterns can be detected approximately 10–14 days before relapse, with strong predictive performance achieved using gradient boosting models. Key features include mobility entropy and sleep irregularity, which consistently emerged as significant indicators. This work contributes to the growing field of digital phenotyping and computational psychiatry by providing a scalable framework for continuous mental health monitoring and early intervention. This manuscript is currently under review at JMIR Formative Resear