Mental illness contributes significantly to the global burden of mental disorders. Unlike previous research, this study employs a data-driven exploratory research approach to explore the complex factors underlying mental health well-being in the U.S. cities. Specifically, we use advanced statistical learning algorithms to model and predict the mental health effects of the built environment and socio-economic conditions among adults, controlling for preclinical conditions and behavioral factors. We establish our framework using the metropolitan census tract regions of the five states - Nevada, Oregon, Idaho, Utah, and Wyoming - that generally top the list of poor mental health ranking in the U.S. Our results show that the ensemble tree-based models best capture the complex nexus among mental health, built environment and socio-economic conditions. The analysis conducted herein suggests that mental health outcomes among adults are affected by decline in neighborhood characteristics (i.e., high vacancy rate and long duration in vacancy), lack of health insurance and high incidence of poverty within the metropolitan areas studied in the five states. Policy efforts and conversations around mental health issues must confront the often non-linear interaction of built environment and socio-economic factors that affect mental health outcomes in cities.