Classification of Mental Health Level of Students Using SMOTE and Soft Voting Ensemble Classifier and the DASS-21 Profile

Abstract

This study proposes a comprehensive approach to address the rise in mental health problems among college students. It leverages the Synthetic Minority Over-sampling Technique (SMOTE) to address the class imbalance in the dataset and employs a Voting Ensemble with soft voting to combine several base algorithms (Logistic Regression, Random Forest, Gradient Boosting, and XGBoost/SVM) for accurate prediction of mental health levels (normal, mild, moderate, severe, very severe). Feature importance-based feature selection using Random Forest is utilized to eliminate less relevant attributes. The model evaluation includes accuracy, precision, recall, F1-score, and confusion matrix analysis. The results demonstrate that the ensemble approach improves stability and accuracy compared to individual models. Notably, the application of SMOTE led to significant performance improvements, with classification accuracies reaching up to 100% for the Random Forest model. These findings support the use of ensemble learning and SMOTE for developing effective college student mental health screening systems, ultimately enabling timely intervention and support.