A Predictive Modeling Framework for Early Behavioral Intervention and Privacy-Preserving Digital Care

• Abey city

  Lautech

  Author

The escalating global burden of mental health disorders among higher education students has exposed critical limitations in traditional reactive care models, including workforce shortages, temporal lag in detection, and barriers to help-seeking. While artificial intelligence (AI) has emerged as a promising tool for enhancing psychological distress detection, existing approaches face persistent challenges: intrusive data collection, limited explainability, privacy violations, and the risk of prematurely medicalizing routine behavioral variation. This study addresses these gaps by proposing and validating the Generative Semantic Intermediary Framework (GSIF), a privacy-preserving predictive modeling architecture that integrates multimodal behavioral signals—including learning management system engagement, smartphone sensor data, and voice features—through a two-stage large language model translation process. The framework achieves early warning detection with 89.4% accuracy (AUC = 0.91), representing a 23.7% improvement over unimodal screening approaches, while reducing false-positive burden through constrained review prioritization and human-in-the-loop verification. Our findings demonstrate that semantic translation of behavioral indicators into clinically reviewable descriptors, combined with graph-personalized federated learning for privacy preservation, offers a technically rigorous and ethically aligned pathway for continuous mental health monitoring in higher education. The framework contributes a replicable digital health architecture that balances predictive performance with data minimization, role-bounded access, and transparent escalation thresholds.

• PDF

Copyright (c) 2026 Abey city (Author)

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