GeoAI-Driven Housing Affordability Analysis: Predicting Rent Burden at the Census Block Level Across New York City

This dataset supports the replication of a GeoAI-driven housing affordability analysis predicting rent burden at the census block level across New York City. The package contains all analytical inputs, model artifacts, spatial outputs, and publication figures produced from 2020 American Community Survey five-year estimates for 35,099 census blocks spanning all five NYC boroughs.
The core dataset (NYC_Blocks_Renamed.csv) contains 37,816 census blocks with 394 cleaned ACS attribute columns covering income, poverty, housing stock, demographics, built environment, employment, transport, and environmental quality. The results file (NYC_Blocks_Affordability_Results.csv) provides block-level predictions from Random Forest (R²=0.684) and XGBoost (R²=0.756) regressors, Local Indicators of Spatial Association (LISA) cluster assignments, and K-means affordability typology labels.
Key findings include: 80% of NYC census blocks exceed the HUD 30% cost-burden threshold; the citywide median rent burden rate is 49.6%; Global Moran's I = 0.5686 (p<0.001) confirms strong spatial clustering; and Spatial Error Model lambda = 0.681 indicates that spatially structured unobserved factors are the primary driver of rent burden concentration. SHAP analysis identifies Rent-to-Income Ratio, Median Household Income, and Renter Rate as dominant predictors.
Model artifacts (serialized XGBoost, Random Forest, LASSO-selected feature list, and imputer) enable direct replication of predictions without re-training. Four publication-quality maps are included covering rent burden rates, LISA hotspot clusters, and K-means affordability typologies. An interactive prediction tool is permanently deployed at https://huggingface.co/spaces/omansour2222/nyc-rent-burden-predictor, enabling scenario-based analysis without specialist software.