Affiliations and Data Partners
At Pairwell Research Lab
Faculty from multiple departments at Algoma University are developing an automated monitoring system to track student well-being across various domains, such as mental health, housing, and social support, while ensuring privacy and confidentiality. The system will provide real-time data for early intervention, incorporate AI analytics, and reduce stigma by normalizing mental health care, ultimately informing future interventions and curriculum development.
Publications
Privacy-
aware student mental health and wellness monitoring system
Authors: Ajmery Sultana, Wenjun Lin, M. Nasir, Miguel A. Garcia-Ruiz, Teryn Bruni, Walter Chan
International Conference on Mental Health and Addictions, April 2024.
Student wellness challenges—including food insecurity, housing crises, and mental distress—impair academic and social success, yet students typically delay seeking help until problems become acute. Early identification of at-risk students is essential but hindered by inadequate measurement tools. This study proposes an automated monitoring system to assess multiple wellbeing domains (mental health, physical health, housing, food security, social support) while maintaining privacy standards. The approach involves stakeholder interviews and surveys across three Algoma University campuses to capture diverse student needs, followed by AI-driven predictive modeling. The system will provide regular status updates, enabling proactive interventions and supporting wellbeing specialists in identifying vulnerable students early.
Cost-Effective Predictive Modeling for Student Mental Health Using Readily Available Data
Authors: Shahroz Abbas, Filip Al-Hamadani, Ajmery Sultana, M. Nasir, Miguel A. Garcia-Ruiz, Wenjun Lin
IEEE International Conference on Computing and Machine
Intelligence (ICMI), April 2025.
Post-secondary student mental health is a pressing public health concern, with high rates of psychological distress and suicidal ideation on campuses. While predictive modeling using large surveys can assess students’ mental states, collecting such data is expensive and time-consuming, often delaying proactive interventions. This paper investigates whether readily available institutional data can predict student mental health with comparable accuracy. Results demonstrate that accessible data combined with neural network models achieves competitive predictive performance relative to studies using costly survey-based approaches, offering a more scalable and timely solution for institutions to support student wellbeing.
Privacy-Preserving Machine Learning for Mental Health Prediction Using Homomorphic Encryption
S. Abbas, F. Al-Hamadani, A. Sultana, M. Nasir, M. Garcia-Ruiz, and W. Lin
Proc. IEEE 4th Int. Conf. Comput. Mach. Intell. (ICMI), Central Michigan, USA, Apr. 5–6, 2025.
Student mental health issues like stress, anxiety, and depression significantly impact academic performance, yet existing machine learning prediction systems process sensitive data in plaintext, risking privacy breaches. While homomorphic encryption (HE) enables secure ML, current implementations face impractical computational overhead. This paper introduces a privacy-preserving predictive model using logistic regression trained on encrypted data via TenSEAL, employing leveled fully homomorphic encryption with a quadratic sigmoid approximation for compatibility. A comprehensive efficiency analysis evaluates RAM usage and training time across polynomial-modulus degrees. Results show the encrypted model achieves 84% accuracy (versus 96% unencrypted) with scalable resource consumption, advancing feasible privacy-preserving ML for sensitive mental health applications.
A Secure and Scalable Quantum-Resisant Federated Learning Service for Sensitive Data Analytics
S. Abbas, Ajmery Sultana, M. Nasir, M. Garcia-Ruiz, and W. Lin, IEEE Transaction on Service Computing, March 2026
With the advances of quantum computing the security of existing cryptographic frameworks is increasingly at risk. Accordingly, in the present study, we investigate the integration of post-quantum cryptographic algorithms into Hyperledger Fabric, a blockchain framework, to safeguard it against emerging quantum threats. To this end, a modified Cryptogen tool was developed to generate X.509 certificates with both classical and post-quantum cryptographic keys. Furthermore, using tools like Hyperledger Caliper and Prometheus for empirical analysis, we demonstrate that this hybrid approach effectively strengthens security without affecting system performance. These results not only improve the security of Hyperledger Fabric, but also offer a practical guide for adding post-quantum cryptography to blockchain technologies.
