Explainable machine learning for urban housing prices: Policy relevant diagnostics across three cities

• Explainable ML is a practical diagnostic complement to hedonic pricing models. • Tree-based models improve out-of-sample R 2 by 0.10–0.35 over linear baselines. • SHAP yields spatially explicit, dollar-valued insights for non-linear thresholds. • Continuous distance gradients reveal income-differentiated proximity premiums. • Diagnostics inform TOD zoning, appraisals, and targeted anti-displacement policies.