The quality of pavement networks is greatly affected by different distresses. These distresses appear in many forms, such as cracking, potholes, rutting and different types of deformation. As a result, to ensure effective pavement management, accurate modeling of these different distresses has become essential. Moreover, machine learning models have shown great potential in modeling pavement performance in recent years. The objective of this research is to develop machine learning models for modeling key parameters of pavement distress, specifically the International Roughness Index (IRI), fatigue and longitudinal cracking. Data for this investigation were extracted from the Long-Term Pavement Performance (LTPP) database, with a focus on areas exhibiting environmental conditions similar to those in Egypt. By doing so, the models would be applicable to Egyptian settings. The dataset comprised of 8537 datapoints on 221 different pavement sections. The variables collected include IRI, temperature, precipitation, Equivalent Single Axle Loads (ESALs), pavement age, time since last maintenance, asphalt concrete layer thickness, average asphalt content, bulk specific gravity, granular base thickness, percentage of fatigue cracking, and percentage of longitudinal cracking. Six machine learning algorithms were used for modeling each output variable: XGBoost, Random Forest, K-Nearest Neighbors (KNN), Bayesian Regression, Ridge Regression, and Decision Trees. Model performance was assessed using Mean Absolute Error (MAE) and R2 as evaluation metrics. Comparative analysis revealed that the XGBoost algorithm demonstrated superior performance in modeling all three output variables. The results showed a MAE of 0.17 and R2 of 0.729 for modeling IRI. For modeling fatigue cracking and longitudinal cracking, the model produced a MAE of 4.92% and 2.96%, respectively, with an R2 of 0.672 and 0.692 respectively. The findings are significant for many reasons. Firstly, they offer a framework for modeling pavement distress parameters, which is crucial for effective pavement management and maintenance strategies. Secondly, the study confirms the efficacy of machine learning algorithms in modeling pavement performance indicators, especially when using ensemble models. Lastly, the exceptional performance of the XGBoost algorithm indicates its reliability as a tool for both future research and practical applications in pavement management. Importantly, the models are tailored to be applicable in Egypt, providing a data-driven approach to improve the quality of road infrastructure in the region.
School of Sciences and Engineering
Construction Engineering Department
MS in Construction Engineering
Committee Member 1
Committee Member 2
Sherif El Badawy
Committee Member 3
Institutional Review Board (IRB) Approval
Not necessary for this item
(2024).Prediction of Distresses in Pavement Networks: A Machine Learning Approach [Master's Thesis, the American University in Cairo]. AUC Knowledge Fountain.
Kotb, Mahmoud. Prediction of Distresses in Pavement Networks: A Machine Learning Approach. 2024. American University in Cairo, Master's Thesis. AUC Knowledge Fountain.