Abstract: Mental health challenges among college students have reached epidemic proportions, with approximately 40% of students experiencing significant psychological distress during their university years. Traditional reactive approaches to student mental health support have proven inadequate in addressing the scale and complexity of this crisis affecting millions of students globally. This study presents an Enhanced Adaptive K-Nearest Neighbor (EAKNN) algorithm for early detection of psychological issues in college students through intelligent analysis of behavioral patterns and temporal changes. The proposed system learns individual baseline behaviors, adapts to personal patterns, and identifies concerning deviations that may indicate developing mental health concerns before they escalate. Experimental validation with 200 college students over six months demonstrated that EAKNN achieved 92.5% overall accuracy, 92.7% sensitivity, and 91.1% precision in identifying at-risk students an average of 3.7 weeks before traditional screening methods. The algorithm incorporates temporal weighting, adaptive feature importance, and explicit uncertainty quantification to provide personalized, interpretable assessments tailored to individual circumstances. Statistical analysis confirmed significant improvements over baseline methods including Random Forest (p < 0.001, Cohen's d = 0.82). This research demonstrates that personalized, adaptive machine learning approaches can transform mental health support from reactive crisis management to proactive early intervention, potentially improving outcomes for thousands of students while optimizing limited counseling resources.
Keywords: Mental Health Detection, Machine Learning, EAKNN Algorithm, College Students, Early Detection, Behavioral Analysis, Predictive Analytics, Student Wellbeing.

Downloads: | DOI: 10.17148/IJIREEICE.2025.131109