Prediction of depression in university students using five-fold cross-validation and machine learning models

Abstract

Depression among university students has become a significant public health concern due to increasing academic pressure, financial instability, emotional stress, and social challenges. Early prediction of depression can help institutions implement timely interventions and improve student well-being. The study implements and compares four different classification algorithms: Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN). The analysis incorporates data augmentation techniques, handles class imbalance through SMOTE (Synthetic Minority Over-sampling Technique), and provides extensive model evaluation metrics, including standard train-test split evaluation and rigorous 5-fold cross-validation with aggregated confusion matrices. Data preprocessing involved handling missing values, normalization, label encoding, and feature selection. Model performance was evaluated using accuracy, precision, recall, F1-score, and Receiver Operating Characteristic Area Under Curve (ROC-AUC). Experimental results demonstrated that the Random Forest model achieves superior performance, with 91.5% accuracy and 0.904 cross-validation accuracy, significantly outperforming other architectures for predicting depression among students. The findings indicate that machine learning approaches can provide reliable tools for early depression risk identification in higher education institutions.