Mental health disparities are significantly influenced by social determinants such as income level, education, employment status, access to healthcare, and environmental factors. Traditional methods for identifying at-risk populations often rely on statistical models that fail to capture the complex, nonlinear interactions between these determinants. In this study, we propose a machine learning-driven approach to uncover mental health disparities using a combination of Neural Network Modeling, Reinforcement Learning, and Random Forest. We leverage publicly available datasets, including those from the Centers for Disease Control and Prevention (CDC) and the U.S. Census Bureau, to extract critical socioeconomic and behavioral health indicators. Our methodology involves employing Random Forest for feature selection, ensuring the most influential social determinants are identified, followed by Neural Networks to model complex relationships and predict mental health outcomes. Additionally, Reinforcement Learning is integrated to optimize policy recommendations by simulating interventions that could mitigate disparities. The models are evaluated based on accuracy, precision, recall, and fairness metrics to assess their predictive power and ability to provide equitable insights. The results demonstrate that our hybrid machine learning approach significantly improves mental health outcome predictions compared to traditional statistical models. Moreover, the reinforcement learning framework offers adaptive intervention strategies that could assist policymakers and healthcare professionals in addressing disparities more effectively. Our findings underscore the potential of AI-driven analytics in promoting mental health equity and guiding data-driven policy decisions.