Analisis Prediksi Risiko Depresi pada Remaja Menggunakan Random Forest dan Logistic Regression Berdasarkan Pola Penggunaan Media Sosial

Abstract

Depression is a common mental health disorder among students and can negatively affect psychological well-being and academic performance. This study aims to develop and compare depression classification models using Logistic Regression and Random Forest algorithms. The dataset consisted of 1,200 records with an imbalanced class distribution. The research process included data cleaning, categorical feature transformation using One-Hot Encoding, class balancing through the Synthetic Minority Oversampling Technique (SMOTE), data splitting into training and testing sets, and model training and evaluation. Model performance was assessed using accuracy, precision, recall, F1-score, and ROC-AUC metrics. The results indicate that Random Forest outperformed Logistic Regression, achieving an accuracy of 99.79%, precision of 100%, recall of 99.57%, F1-score of 99.79%, and ROC-AUC of 100%. In comparison, Logistic Regression achieved an accuracy of 96.37%, precision of 97.38%, recall of 95.30%, F1-score of 96.33%, and ROC-AUC of 99.60%. Feature importance analysis revealed that sleep duration, anxiety level, stress level, and daily social media usage were the most influential factors associated with depression. These findings suggest that Random Forest is an effective approach for supporting the early detection of depression based on individual characteristics and behavioral patterns among adolescents.