Student Mental Health Prediction using Maching Learning Techniques

Abstract

This research delivers a machine learning prediction system which tracks student stress levels and smartphone addiction behaviors because these control mental health and educational outcomes. The rising addiction to smartphones directly links to enhanced stress levels while simultaneously causing students to perform worse at school and resulting in diminished mental well-being throughout the era of advanced digital usage. A study examined over 1000 pupil behavioral and physiologic and smartphone usage data with ten machine learning models comprising Naïve Bayes, Decision Tree, XGBoost, LightGBM, Gradient Boosting, SVM, Random Forest and KNN and Logistic Regression and AdaBoost. Results from accuracy performance assessment show Logistic Regression achieved 0.8066, Random Forest hit 0.9811, SVM reached 0.8396 and Gradient Boosting exceeded 0.9858, KNN settled at 0.8538, Decision Tree reached 0.9906, AdaBoost reached 0.7689, XGBoost attained 0.9906, LightGBM touched 0.9717 and Naïve Bayes ended at 0.7925. Stress prediction through CatBoost stands out as the most effective model since it produces the lowest MSE alongside the highest R² when compared to Linear Regression for addiction forecasting. Researchers determined the importance of early mental health interventions through their analysis which showed substantial connections between stress levels and smartphone usage conduct and educational results. The study proves that machine learning serves as an effective mental health prediction system so educational institutions can create specific stress management and addition prevention approaches. The research outcomes support real-time data monitoring systems through their dual role of precise forecasting and the development of superior mental health programs.