Explainable AI untuk Deteksi Tingkat Depresi Mahasiswa Berbasis Faktor Akademik dan Sosial: Pendekatan SHAP

Abstract

Depression is one of the most prevalent mental health disorders among university students and significantly affects academic performance and psychological well-being. Machine learning models have been widely applied for early depression risk detection, yet their black-box nature limits their use as a basis for intervention by counselors and education policymakers. This study develops an interpretable model for detecting student depression levels based on academic and social factors using an Explainable AI (XAI) approach, specifically SHapley Additive exPlanations (SHAP). The dataset consists of 502 student respondents with 10 features, including academic pressure, study satisfaction, sleep duration, dietary habits, suicidal thoughts, study hours, financial stress, family history of mental illness, age, and gender. Three classification models were compared: Logistic Regression, Random Forest, and XGBoost. Logistic Regression achieved the highest accuracy at 98.02%, followed by XGBoost at 94.06% and Random Forest at 93.07%. SHAP interpretation on the XGBoost model revealed that suicidal thoughts, academic pressure, age, study satisfaction, and financial stress were the five most influential predictors of depression level. These findings provide an empirical basis for higher education institutions to design early-detection systems and targeted, transparently accountable student mental health support policies.