Mental Health Prediction Using Ensemble Learning Approaches with Rebalancing Technique

Mental health disorder like anxiety, depression, and stress are common among university students and can affect their well-being and academic performance. Early and precise detection of mental health disorders is essential for therapy and control. The existing prediction approaches, which depend on single classifier are often insufficient due to data complexity and class imbalance, resulting in unreliable predicitions of mental health disorder. This study aims to propose an ensemble learning approach for classifying the different categories of mental health disorder using personal information and responses of mental health surveys (Anxiety, Depression and Stress). This study evaluates the performance of three ensemble learning classifiers, including Gradient Boosting (Boosting), Random Forest (Bagging) and K-Nearest Neighbour (Stacking) on the university student’s dataset, consists of student information classified into five different categories: Normal, Mild, Moderate, Severe and Extremely Severe. Gradient Boosting demonstrated the highest accuracy and G-Mean for Anxiety (83.8%) and Depression (83.8%) classification, while Random Forest outperformed in Stress classification (77.3%), achieving the highest recall and precision. Gradient Boosting exhibited strong performance in accurately differentiating between various levels of mental health disorder, as indicated by the results of the receiver operating characteristic (ROC) curve. Future research will focus on integrating additional behavioral datasets to strengthen the model’s resilience and expand its diagnostic capabilities for a wider range of mental health disorders.