The Impact Of Crime And Spatial Features on Housing Prices: An Explainable Ai Approach In New York City

This study presents a machine learning–based framework that integrates spatially linked real estate transaction data and crime records to predict residential property prices in the New York City (NYC) housing market over the period 2010–2024. The motivation stems from the limitation of traditional valuation approaches, which primarily emphasize structural property attributes while inadequately accounting for environmental and neighborhood-level factors. To address this gap, crime data from the NYC Police Department and real estate sales records from the NYC Department of Finance were spatially combined using Neighborhood Tabulation Areas.The resulting dataset comprises annual neighborhood-level residential sales statistics and crime data categorized into nine major types. Seven machine learning algorithms—Linear Regression, Ridge, LASSO, Random Forest, XGBoost, LightGBM, and CatBoost—were trained and evaluated using 5-fold cross-validation. Model performance was assessed using R², RMSE, MAE, and MAPE metrics.The results demonstrate that the CatBoost model achieved the highest and most stable predictive performance, with an R² value of approximately 0.75. Further analysis indicates that neighborhood characteristics are the most influential predictors of housing prices, while crime-related variables also make a meaningful contribution. Model interpretability was enhanced through SHapley Additive exPlanations (SHAP), which quantified the marginal effects of both spatial and crime-related features. Overall, the findings confirm that incorporating spatial crime information improves predictive accuracy and highlight the importance of security as a key determinant of housing prices in heterogeneous urban environments such as NYC.