This study examines the application of machine learning models for predicting risks in mental health issues and shows a comparative analysis of various algorithms focusing on K-Nearest Neighbors (KNN), Logistic Regression (LR), Decision Trees (DT), Random Forest Classifiers (RFC), Ada Boost Classifier, and Gradient Boosting Classifier. The SMOTEENN approach was applied or educe class imbalance in the dataset. This technique enhances the balance of the dataset and also the whole predictive performance of the models. Hyperparameter tuning optimized model parameters, and significant results were obtained for enhancing the accuracy and F1 scores across all models. Applying L1 and L2 regularization to reduce over fitting for better reliability of models revealed that the Random Forest Classifier outperformed other algorithms with a near accuracy of about 86.66%. These findings highlight the possible role of machine learning in early detection and proactive management of mental health risks. As such, data-driven approaches are likely to give new insights to mental health professionals. The study is, therefore, a valuable contribution to the growing body of literature on mental health analytics and underscores the importance of robust methodologies in predicting outcomes for mental health