Integrating Multimodal Artificial Intelligence with Passive Sensing for Continuous Youth Mental Health Monitoring

• Abey city

  Lautech

  Author

Adolescent mental health crises have escalated globally, yet traditional screening methods remain episodic, subjective, and滞后, failing to capture dynamic symptom fluctuations. Existing digital monitoring tools largely rely on active self-reports, suffering from low adherence and ecological validity gaps. This study addresses the critical need for a continuous, objective, and privacy-preserving framework by integrating multimodal artificial intelligence with passive sensing data from smartphones and wearable devices. Using a design-based research methodology, we collected passive digital biomarkers (accelerometry, touch interactions, keyboard dynamics, ambient light, and location variance) from 210 secondary school students (aged 14–17) over 12 weeks, alongside weekly clinical assessments (PHQ-9 and GAD-7). We developed a hybrid deep learning architecture combining a temporal convolutional network (TCN) with a cross-modal attention mechanism, achieving a predictive accuracy of 89.4% (F1-score = 0.87) for detecting clinically meaningful symptom increases 5–7 days prior to self-reported changes. The framework significantly outperformed single-modality baselines (best baseline accuracy = 74.2%, p < 0.001). Key predictive features included sleep onset variability, typing latency, and location entropy. This research provides a validated, replicable framework for predictive intervention, enabling school counselors and digital behavioral health systems to deliver timely, targeted support. The findings have profound implications for integrating continuous monitoring into secondary education without disrupting daily routines.

• PDF

Copyright (c) 2026 Abey city (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.