A Multimodal AI–IoT Framework for Monitoring Student Well-Being in Hybrid and Transnational Learning Environments

Abstract

Problem Statement: International students often face challenges to their well-being as they adjust to new cultural, academic, and social environments. These challenges can be increased in hybrid learning settings, where students must balance on-campus participation with significant online study. Feelings of isolation, academic pressure, disrupted routines, and difficulties in building local support networks are common. Universities typically lack continuous insight into how these stressors appear in students’ daily behaviours. This study examines how everyday digital and physical activity patterns captured through unobtrusive sensing can help institutions better understand and support the well-being of international students. Methods: Using the publicly available StudentLife dataset as a proxy for the behavioural data students naturally produce, we analyse four key streams: physical activity, sleep estimation, phone usage, and online course engagement. These signals are incorporated into a conceptual IoT-enabled smart campus solution based on a novel architecture, where mobile sensing and environmental context form an integrated layer for well-being monitoring. We extract daily behavioural features and link them with PHQ-9 mental health scores to explore how fluctuations in routine behaviours correspond to changes in mood and well-being. Machine learning models, including Random Forest, Gradient Boosting, and Decision Tree, are used to identify behavioural patterns that align with elevated stress. Results: The analysis shows that irregular sleep patterns, reduced physical activity, increased phone usage, and fluctuations in online engagement frequently coincide with higher PHQ-9 scores, indicating a low level of mental health well-being. These behavioural changes reflect patterns commonly reported by international students during periods of adjustment or academic pressure. Significantly, some behavioural shifts occur before increases in PHQ-9 symptoms, indicating potential opportunities for early, supportive intervention before well-being declines become severe. Significance: This work demonstrates how unobtrusive behavioural sensing, embedded within an IoT-enabled innovative campus framework, can provide universities with empathetic, real-time insights into the well-being of students, especially international students who are highly vulnerable. The study emphasises how institutions can use these insights to offer timely check-ins, culturally responsive support, and personalised resources. The findings contribute to a more human-centred approach to supporting the well-being of international students, helping universities create inclusive and supportive learning environments for a diverse global student population.