This study addresses income inequality in New York City (NYC) by developing a Socio-economic & Spatial-Information-GP (SSIG) model to estimate district-based income. SSIG incorporates socio-economic data with spatial information in a Gaussian Process (GP) model, improving accuracy compared to spatial-only GP models. SHapley Additive exPlanations (SHAP) analysis reveals the relative contributions of socio-economic variables. Results show SSIG's superior accuracy in estimating per-capita and median household income at Tract and ZIP levels. SHAP analysis highlights the predictive power of higher education (bachelor's and postgraduate degrees) on income, while also indicating persistent income inequality due to race and sex, with race having a stronger effect. An ablation study confirms socio-economic information's greater predictive power at the ZIP level. These findings have implications for policy-making, suggesting a focus on addressing income disparities based on race and sex, while expanding higher education opportunities in lower-income districts.

Ruiqiao Bai, Jacqueline C. K. Lam, Victor O. K. Li