PREDICTING OF MENTAL HEALTH With MACHINE LEARNING ALGORITHMS

Abstract

This project presents a comprehensive mental health prediction system utilizing machine learning algorithms to identify individuals at potential risk based on diverse personal and behavioral attributes. The system was developed and evaluated using three different supervised learning models: Support Vector Machine (SVM), Logistic Regression, and Random Forest, to determine the most effective approach for high-accuracy mental health risk assessment. Extensive experiments were conducted on a labeled dataset, with features pre-processed and normalized to enhance model performance and generalizability. Among the tested algorithms, the Random Forest classifier demonstrated superior performance, achieving an overall accuracy of 98.9%, a macro precision of 97.8%, a macro recall of 98.6%, and a macro F1-score of 98.2%, while maintaining efficient computational requirements and robust handling of feature interactions. The SVM model, while delivering solid results with an accuracy of 94.2%, exhibited longer training times and sensitivity to parameter tuning. Logistic Regression, despite its simplicity and interpretability, achieved comparatively lower performance with an accuracy of 91.7%. These findings highlight Random Forest as the most effective model, balancing predictive power and computational efficiency for mental health risk detection. The entire system was implemented using Python and popular machine learning libraries, facilitating scalable deployment in digital health applications. Future directions may involve integrating additional behavioral and physiological data, deploying models in real-time web or mobile applications, and exploring explainable AI techniques to enhance transparency and trust in mental health predictions.