Mapping Heterogeneity in Psychological Risk Among University Students Using Explainable Machine Learning

Area of Science:

• Computational Mental Health
• Artificial Intelligence in Psychology
• Higher Education Student Well-being

Background:

• Student mental health is a critical post-pandemic issue in higher education.
• Conventional assessments often overlook the heterogeneity of at-risk student populations, limiting intervention effectiveness.
• There is a need for advanced methods to understand nuanced psychological risk mechanisms.

Purpose of the Study:

• To develop a novel computational framework integrating explainable artificial intelligence (XAI) and unsupervised learning.
• To decode the latent heterogeneity of psychological risk mechanisms in students.
• To establish a foundation for precision interventions targeting specific risk drivers.

Main Methods:

• A "predict-explain-discover" pipeline was developed using TreeSHAP and Gaussian Mixture Models.
• Identified distinct risk subtypes based on a 2556-dimensional feature space (lexical, linguistic, affective indicators).
• Sensitivity analysis using top-20 core features validated the structural stability of identified mechanisms.

Main Results:

• Identified three theoretically-grounded risk subtypes: academically-driven (28.46%), socio-emotional (43.85%), and internal regulatory (27.69%).
• Subtypes were validated as anchored in primary decision drivers, not high-dimensional noise.
• Demonstrated transformation of black-box classifiers into diagnostic tools.

Conclusions:

• The framework bridges predictive accuracy and mechanistic understanding in computational mental health.
• Findings align with Research Domain Criteria (RDoC), supporting precision interventions.
• Advances mechanism-based subtyping for personalized student support in higher education.