AI-DRIVEN CLIMATE-RESILIENT PAVEMENT DETERIORATION PREDICTION FOR SUSTAINABLE ROAD INFRASTRUCTURE IN PAKISTAN

Abstract

Climate change, increasing traffic loading, and aging road infrastructure create growing challenges for pavement management in Pakistan. This study developed an artificial intelligence (AI)-driven framework for predicting pavement deterioration by integrating pavement condition, traffic loading, maintenance history, and climatic variables, including temperature, precipitation, and flood exposure. A quantitative longitudinal design was used at the road-section level. Random Forest, Support Vector Regression, Gradient Boosting, XGBoost, and Artificial Neural Network models were evaluated using R², RMSE, MAE, and MAPE. The reported analysis identified XGBoost as the strongest-performing model and showed that the climate-sensitive model outperformed the conventional pavement-and-traffic model. Pavement age, traffic loading, previous pavement condition, maximum temperature, flood exposure, and maintenance interval were important predictors. SHAP-based explainability further indicated the direction and contribution of major predictors. The framework supports earlier identification of vulnerable road sections and a transition from reactive maintenance toward predictive, climate-resilient, and sustainability-oriented pavement management.