NeuroSym-MHIR: A Neuro-Symbolic Hybrid Intelligence Framework for Multi-Stage Mental Health Risk Stratification, Behavioral Phenotyping, and Personalized Intervention Recommendation

Abstract

Increased levels of stress, anxiety, and depression among university students have put mental health care under the spotlight for universities around the globe. This study examines the potential for using AI and/or ML methods to produce data-driven individualized intervention suggestions from readily accessible student data. We performed exploratory data analysis, correlation analysis, supervised risk-tier classification with the aid of a Random Forest model, and unsupervised classification with K-Means clustering to discover the underlying student behavior profiles in the cross-sectional survey data consisting of 7,022 university students on demographic, academic, lifestyle, psychological (stress, depression, anxiety scores), sleep quality, physical activity, diet, social support, substance use, financial stress, and use of counselling services. A mental health risk index was carefully constructed as a composite of the three psychological scales and a rule-based recommendation engine was built to map each behavioral cluster and mental health risk level to a specific bundle of interventions (e.g., sleep-hygiene coaching, peer support referral, counselling escalation, financial-aid signposting). Reported measures of stress, depression, and anxiety are only slightly correlated with lifestyle and demographic variables in this data set (with |r| < 0.03 for most variables being tested), and a Random Forest classifier trained using the lifestyle and academic features to predict the stress degree to within a risk tier had modest accuracy (39.7%, macro-F1 = 0.37), just better than chance. Analyzing lifestyle and help-seeking variables in clusters resulted in four interpretable behavioral personas, which were significantly different in reported symptom severity, sleep quality, and help-seeking, but not substance use.