Loneliness and depression are interrelated mental health issues affecting students well-being. Using passive sensing data provides a novel approach to examine the granular behavioural indicators differentiating loneliness and depression, and the mediators in their relationship. This study aimed to investigate associations between behavioural features and loneliness and depression among students, exploring the complex relationships between these mental health conditions and associated behaviours. This study combined regression analysis, mediation analysis, and machine learning analysis to explore relationships between behavioural features, loneliness, and depression using passive sensing data, capturing daily life behaviours such as physical activity, phone usage, sleep patterns, and social interactions. Results revealed significant associations between behavioural features and loneliness and depression, emphasizing their interconnected nature. Increased activity and sleep duration were identified as protective factors. Distinct behavioural features for each condition were also found. Mediation analysis highlighted significant indirect effects in the relationship between loneliness and depression. The XGBoost model achieved the highest accuracy in predicting these conditions. This study demonstrated the importance of using passive sensing data and a multi-method approach to understand the complex relationship between loneliness, depression, and associated behaviours. Identifying specific behavioural features and mediators contributes to a deeper understanding of factors influencing loneliness and depression among students. This comprehensive perspective emphasizes the importance of interdisciplinary collaboration for a more nuanced understanding of complex human experiences.