CASTLE: A Multi-Modal Educational Data Fusion Framework for Student Mental Health Detection using MOON Network Embedding and Deep Neural Networks

Abstract

Mental health issues among university students pose significant risks, including depression, self-harm, and other severe consequences. However, many affected individuals are unaware of their condition and do not seek professional help. Early identification of mental health concerns is crucial, yet challenging due to the unstructured and multi-modal nature of data generated in academic and social settings. To address this, we introduce CASTLE (Comprehensive Analysis of Student Traits and Learning Environment), a novel deep learning framework that leverages multi-modal educational data fusion for proactive mental health detection. Our approach integrates diverse information sources, including social interactions, academic performance, physical attributes, and demographic variables, to construct a comprehensive representation of students' well-being. A multi-perspective social network embedding technique, MOON (Multi-view SOcial NetwOrk EmbeddiNg), is employed to model heterogeneous social connections and eliminate redundant information. To counteract data imbalance, the Synthetic Minority Oversampling Technique (SMOTE) is utilized, enhancing model robustness. Finally, a deep neural network (DNN) is trained for accurate classification of mental health conditions. Experimental evaluations demonstrate that CASTLE achieves a recall of 84.5% and an F1-score of 73.6%, significantly outperforming state-of-the-art baselines. This study highlights the potential of AI-driven solutions in fostering mental well-being and providing early interventions in educational environments.