Machine Learning Approaches for Predicting Suicide Risk in Adolescents Presenting to Emergency Services in Lagos State

Machine learning techniques offer promising capabilities for improving suicide risk prediction beyond traditional clinical assessment tools, with significant implications for emergency psychiatric triage. This study develops and validates machine learning models for predicting suicide risk in adolescents presenting to emergency services in Lagos State, South West Nigeria. A retrospective-prospective mixed design will compile a dataset from 300 adolescent emergency presentations involving suicidal behavior across three Lagos State hospitals over three years. Candidate predictor variables will include depression scores (PHQ-A), history of previous attempts, substance use, impulsivity ratings, and demographic features. Random forest, logistic regression, and gradient boosting classifiers will be trained and validated using 10-fold cross-validation and an independent prospective test cohort. Model performance will be evaluated using AUC-ROC, sensitivity, specificity, and positive predictive value. Ethical considerations regarding algorithmic bias, data privacy, and clinical deployment will be addressed systematically. This study is the first to apply machine learning to adolescent suicide risk prediction in a Nigerian emergency context, generating evidence with direct clinical translation potential for Lagos State emergency services. Keywords: machine learning, suicide risk, adolescents, Lagos State, emergency psychiatry