AI-Based Decision Support System for Sustainable Infrastructure Development

Abstract

Sustainable infrastructure development requires integrated assessment of economic, environmental, energy, and social conditions to support evidence-based planning and resource allocation. Existing studies have applied Machine Learning (ML) and Multi-Criteria Decision-Making (MCDM) techniques for sustainability assessment, but limited efforts integrate predictive modelling, explainability, objective weighting, and infrastructure ranking within a unified framework. This study develops an explainable Artificial Intelligence (AI)-based decision-support framework using global development indicators to identify influential factors and assess sustainable infrastructure performance across countries and entities. The framework comprises data preprocessing and screening, Recursive Feature Elimination (RFE), Light Gradient Boosting Machine (LightGBM)-based prediction, SHapley Additive exPlanations (SHAP)-based interpretation, Logarithmic Percentage Change-driven Objective Weighting (LOPCOW), and Evaluation based on Removal Effects of Criteria and Uncertainty-based Normalization and Scoring (ERUNS)-based ranking, followed by Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) validation and sensitivity analysis. RFE identified 10 priority indicators, achieving a validation Root Mean Square Error (RMSE) of 2.34. LightGBM achieved R² = 0.9925. SHAP identified urban electricity access (18.783) and rural electricity access (5.649) as the most influential features. The framework provides an interpretable basis for sustainable infrastructure assessment and supports transparent, data-driven planning and prioritization.