Increased mental health awareness has helped benefit society as a whole. Addressing the needs of individuals and helping improve conditions can help improve the quality of life not just for individuals but for the whole world population. Accurate diagnosis and timely treatment can greatly improve individual outcomes later in life. However, mental health can often be considered a taboo topic, with individuals often avoiding diagnosis due to the associated stigma. This work explores the perspective of using machine learning (ML) to help detect and identify common types of mental disorders based on clinical symptoms. The extreme gradient boosting (XGBoost) model is utilized. However, due to the heavy reliance on hyperparameter selection, an altered implementation of the particle swarm optimization (PSO) algorithm is prepared to help attain desired performance. A comparative analysis against several contemporary metaheuristics is conducted on a publicly available real-world dataset to determine the viability of the suggested technique. The best-performing models attain an accuracy exceeding 97%, suggesting that such a model could be beneficial in real-world applications.