Multimodal Behavioural Data Mining for Mental Health: A Comprehensive Survey on Wearables and Digital Phenotyping

Mental health disorders, particularly depression and anxiety, represent a growing global burden, affecting more than 300 million individuals worldwide and contributing significantly to disability and reduced quality of life. Traditional assessment methods rely heavily on self-reported questionnaires and clinical interviews, which are episodic, subjective, and often fail to capture dynamic behavioral patterns. The emergence of wearable devices and digital phenotyping technologies has enabled continuous, real-time collection of multimodal behavioral data, including physiological signals, activity patterns, sleep metrics, smartphone usage, and social interaction traces. This survey provides a comprehensive review of multimodal behavioral data mining techniques applied to mental health monitoring and prediction. We systematically analyze over recent studies employing machine learning, deep learning, ensemble models, and hybrid data fusion strategies for depression and anxiety detection. Statistical trends indicate that multimodal approaches improve predictive performance by 8–15% compared to unimodal systems, with deep learning architectures such as LSTM and transformer-based models achieving reported accuracies above 85% in controlled datasets. The survey categorizes existing methodologies based on data modalities, feature extraction techniques, fusion strategies (early, late, and hybrid fusion), personalization mechanisms, and evaluation metrics. Furthermore, we examine key challenges, including data heterogeneity, class imbalance, privacy leakage risks, interpretability limitations, and generalization across populations. The paper concludes by outlining future research directions in explainable AI, federated learning, adaptive personalization frameworks, and privacy-preserving multimodal analytics. This survey aims to provide researchers and practitioners with a structured roadmap for advancing data-driven, scalable, and personalized mental health prediction systems using wearable and digital behavioral data.