Mental Health Monitoring in Students: A Classifier-Based Machine Learning Approach

This paper describes a strong machine learning system for monitoring students' mental health using classifier-based predictive modeling. Existing work indicates the application of supervised learning for psychological evaluation; nevertheless, most techniques have limited accuracy, low generalizability, and inappropriate management of unbalanced class distributions, rendering them unsuitable for real-world deployment. To overcome these issues, we present a complete pipeline that includes feature selection, data balancing, and classifier modification to improve detection performance.The dataset utilized, Student Mental Health (source: Kaggle), contains behavioral and psychological markers self-reported by students. Initial preparation includes null-value elimination and label encoding. Outlier samples are deleted using IQR-based detection, and class imbalance is addressed via SMOTE oversampling. The Chi-Square test is used to identify features, which are then reduced dimensionally using Principal Component Analysis (PCA) and MinMax normalization.Six classifiers—Random Forest, Extra Trees, Decision Tree, Logistic Regression, Multi-Layer Perceptron, and XGBoost—are trained using different train-test splits (70:30, 80:20, and 90:10). Metrics like as accuracy, precision, recall, and F1-score are used to measure performance, which is then verified using 10-fold cross-validation. GridSearchCV is used to fine-tune hyperparameters and improve classification performance.The study's findings revealed that the Decision Tree classifier had the greatest accuracy of 99.05% after HyperParameter tuning, beating other models in terms of generalization and consistency. The suggested approach outperforms traditional models and may be implemented in academic settings for proactive evaluation of mental health and planning for interventions.