The Urban Health Risk Mapping project leverages crowdsourced data and machine learning technologies to predict the census tract-level health outcomes for ten major US cities, including Austin, Baltimore, Boston, Dallas, Washington, D.C., Houston, Los Angeles, New York City, San Antonio, and San Francisco. The machine-learning-enabled approach has an advantage over the traditional survey methods in terms of time and cost. The actual health outcomes used in training and testing the models are accessed from the CDC’s 500 Cities Project. The features are created based on three data sources, namely the CDC’s Social Vulnerability Index (SVI) dataset, the EPA’s Smart Location Database (SLD), and the 311 service request datasets accessed from each municipality. Sixty features (i.e., predictor variables) are considered, which characterize the social environment, the physical environment, and the aspects and degrees of neighborhood disorder. A variety of machine learning algorithms are applied and compared, including Ridge Regression, Lasso Regression, Elastic Net, Support Vector Machine, Decision Tree, Random Forest, Extra Trees, and Gradient Boosting. To improve the model performance, the model hyperparameters are fine-tuned using 10-fold cross-validation. Different sets of features are also experimented with.