Abstract
Urban public safety is shaped by the interaction of environmental, infrastructural, and social risks that often overlap across rapidly urbanizing regions. This study develops an integrated framework combining Geographic Information Systems (GIS), Multi-Criteria Decision Analysis (MCDA), and machine learning models to assess composite urban risk. Using spatial datasets on land surface temperature, air quality, drainage, building conditions, demographics, and crime from the selected metropolitan region, the study constructs risk indicators and evaluates cross-domain interactions. Gradient Boosted Decision Trees (GBDT) and Temporal Convolutional Networks (TCN) were trained on harmonized data using k-fold and temporal-block validation. The models achieved strong predictive performance (AUC = 0.91, RMSE = 6.2), confirming robustness. Spatial analysis revealed high-risk clusters along river-adjacent settlements and central transport corridors. Scenario simulations showed that structural upgrades (e.g., drainage enhancement) and green infrastructure reduced composite risks by up to 22-30%, particularly benefiting low-income neighborhoods. Equity-weighted decision analysis further prioritized interventions that improved both safety and resilience. The framework provides a scalable, data-driven approach for diagnosing and mitigating compound urban risks, offering actionable insights for policymakers under resource constraints.