Prediction of Fatigue Life of Asphalt Layers in Desert Environments Using Neural Networks and SHAP Analysis: Al-Kufrh Case Study

Abstract

This study develops a predictive framework using Artificial Neural Networks (ANN) combined with SHAP analysis to estimate the fatigue life of asphalt pavements in Al-Kufra, Libya, a desert environment characterized by extreme temperatures and daily thermal fluctuations. Field data from five roads were analyzed to evaluate the influence of asphalt content and air voids. The ANN model achieved high accuracy (R² = 0.91, MSE = 0.015), showing that asphalt content between 5.5–6.0% significantly improves fatigue resistance, while air voids above 4% reduce service life. SHAP analysis provided transparent interpretation of variable importance, confirming asphalt content as the most supportive factor and air voids as the most detrimental. This research represents the first integrated application in Al-Kufra, bridging local and global scholarship, and offers practical recommendations for sustainable pavement design in hot desert climates.