Machine Learning in Adolescent Mental Health: Advanced Comorbidity Analysis and Text Mining Insights

Dafni Patsiala, Konstantinos Bolias, Fani Passia, Georgios Feretzakis, Athanasios Anastasiou, Yiannis Koumpouros

Background: Justice-involved adolescents exhibit high rates of mental health disorders with complex comorbidity patterns. Understanding these patterns is crucial for developing targeted interventions in this vulnerable population. Methods: We applied multiple machine-learning techniques to electronic records from 124 justice-involved adolescents (11–21 years; mean = 15.7 ± 1.9). Analyses included association rule mining, K-Means clustering with t-SNE visualization, and topic modeling of clinicians’ recommendation notes. Results: Hyperkinetic disorders (F90.0/F90.1) and family-stress factors (Z63.5) together accounted for approximately 45% of all ICD-10 entries. A four-cluster K-Means solution built on age + F-codes alone showed weak separation (silhouette = 0.044), whereas adding Z-codes markedly improved cohesion (silhouette = 0.468) and isolated a distinct hyperkinetic–family-stress subgroup. Association-rule mining returned one robust rule, F81 → F90.0 (support = 0.048, confidence = 0.46, lift = 1.59), underscoring the frequent co-diagnosis of learning and attention-deficit disorders. Topic modeling of clinicians’ recommendation notes recovered five coherent intervention themes—vocational guidance, parent counseling, psycho-education, family psychotherapy, and psychiatric follow-up—which aligned closely with the data-driven clusters. Conclusions: These findings demonstrate how routine clinical data can reveal actionable comorbidity profiles and guide tailored interventions for complex adolescent mental-health presentations.