Explainable Artificial Intelligence for Sustainable Urban Development Decision Support Systems

Abstract

 Artificial Intelligence (AI) systems integrated into urban decision support platforms are becoming more complex and opaquer, posing challenges for contemporary urban governance. AI tools such as land-use zoning, transport demand forecasting and energy consumption optimization have proven highly predictive; however, the opacity of these models has eroded public confidence and regulatory acceptance needed for the responsible use of these tools in critical civic environments. Despite their success at predicting various needs, the opacity of AI models may have compromised user trust and regulatory compliance, making them unsuitable for sensitive civic applications. In this paper, a novel framework named XAI-SUDSS (Explainable Artificial Intelligence for Sustainable Urban Development Decision Support Systems) is introduced that combines the advantages of post-hoc and ante-hoc explainability mechanisms with a multi-objective gradient boosting prediction engine to provide interpretable, auditable and actionable recommendations to urban planners. It is based on a three-tiered explanation architecture: feature attribution (SHAP) for the global, SHAP-based explanation; Local Interpretable Model-agnostic Explanations (LIME) for the instance-level explanation; a counterfactual generation module to address 'what if' planning scenarios. Experiments in four UK cities (Greater London, Greater Manchester, West Midlands, West Yorkshire) using three different datasets show a 31.4% increase in decision acceptance rate for urban planners, a 27.8% decrease in planning approval cycle time and a 94.7% accuracy in zoning recommendations, while keeping 100% compliance with EU AI Act Article 13 transparency requirements. The findings provide a basis for developing XAI-SUDSS as a repeatable and governance-ready framework of explainable AI in sustainable urban development.