Multimodal Mental Health Screening from Student Behavioral and Physical-Activity Indicators

Abstract

The increasing prevalence of mental health disorders among university students has become a pressing global public health concern. Traditional methods for assessing student psychological wellbeing rely predominantly on self-reported questionnaires and clinical interviews, which are often limited by recall bias, low response rates, and the stigma associated with seeking mental health support. This paper explores the development and evaluation of a comprehensive multimodal mental health screening framework that utilizes passive data collected from everyday student behaviors and physical activity indicators. By continuously monitoring digital phenotypes through smartphones and wearable fitness trackers, an unprecedented level of objective, longitudinal data can be harvested. Behavioral indicators include smartphone usage patterns, mobility traces, and digital social interactions, whereas physical activity indicators encompass step counts, heart rate variability, and sleep duration. The proposed framework aligns and fuses these distinct modalities to predict the onset and severity of stress, anxiety, and depression. Utilizing advanced machine learning techniques, specifically ensemble methods and deep neural network architectures, the system extracts critical temporal and spatial features to model the complex underlying relationships between physical routines, digital habits, and psychological states. This research demonstrates that a multimodal approach significantly outperforms unimodal baselines, providing a scalable, non-intrusive, and highly accurate early warning system for educational institutions. The findings emphasize the necessity of integrating diverse data streams to capture a holistic view of student wellbeing, thereby enabling timely interventions and personalized support mechanisms.