Mental health problems encompass a range of conditions that can impact an individual's emotions and behaviors. The conventional methods of mental illness prediction often suffer from the issue of either over-detection or under-detection and the time-consuming manual review process of patients' data during screening sessions. Therefore, this research aims to utilize machine learning techniques to predict mental health problems, complementing the traditional clinical screening and diagnosis process. The proposed models in this project: Logistic Regression, K-Nearest Neighbors, and Random Forest leverage relevant factors from the dataset concerning mental health survey published by Open Source Mental Disorders in 2014 to predict mental health problems. Feature selection and hyperparameter fine-tuning are employed to identify the factors contributing to mental health problems and enhance the performance of the models. The evaluation of these models is measured using accuracy, recall, precision, F1 score, and AUROC. Experimental evaluation results indicated that the Random Forest model utilizing hyperparameters derived from the RandomizedSearchCV method outperforms during model selection using cross-validation. When predicting test set data, it exhibits a good generalization with an accuracy of 83.23%, recall of 89.87%, precision of 78.02%, F1 score of 83.53%, and AUROC of 83.57%.