What question did this study set out to answer?

The research aims to develop machine learning-based models for predicting pavement condition index (PCI) efficiently.

April 30, 2026Open Access

Data Driven Pavement Management: Leveraging Machine Learning for Resilient and Sustainable Pavement Condition Prediction

TRTouqeer Ali RindMehran University of Engineering and Technology SAShiraz AhmedGhulam Ishaq Khan Institute of Engineering Sciences and Technology MJMuhammad Faisal JavedGhulam Ishaq Khan Institute of Engineering Sciences and Technology

Key Points

The research aims to develop machine learning-based models for predicting pavement condition index (PCI) efficiently.
Utilized a dataset from the LaDOTD pavement management system
Employed six machine learning models: artificial neural network, decision trees, random forest, gradient boosting trees, adaptive boosting, and support vector machines
Evaluated model performance using mean squared error, root mean square error, and coefficient of determination.
Random forest and gradient boosting trees showed the highest R2 values at 0.876 and 0.873, respectively
Identified pavement age, precipitation, and initial PCI as the most influential factors
Demonstrated intelligent management practices for pavement agencies to optimize maintenance planning.

Abstract

Infrastructure construction and maintenance is very critical in supporting a robust and sustainable transportation system. The primary goal of transportation infrastructure is to ensure safe and efficient movement. However, aging, traffic loads, and environmental conditions all lead to distress on pavement that necessitates the use of predictive models that aid in efficient maintenance and rehabilitation (M&R). The conventional approach to empirical prediction models of Pavement Condition Index (PCI) has the disadvantage of failing to capture the complex nature of their interdependence and, therefore, is likely to make poor predictions. To address these shortcomings, the study presents state‐of‐the‐art machine learning (ML)–based PCI prediction models that are specific to the climatic factors in Louisiana. This study used a dataset of the LaDOTD pavement management system (PMS) that is complemented with important variables, which include the functional classification (FC), overlay thickness (OT), pavement age, cumulative average truck traffic (CATT), rainfall, and temperature. This research utilizes six ML models, i.e., artificial neural network (ANN), decision trees (DTs), random forest (RF), gradient boosting trees (GBTs), adaptive boosting (AB), and support vector machines (SVMs) in the prediction of PCI values. Statistical performance measurements used to test the performance of the models include Mean Squared error, root mean square error, mean absolute error, and the coefficient of determination ( R 2 ). According to the findings, the best models are RF and the GBTs because they demonstrate the highest value of R 2 (0.876 and 0.873) and the lowest value of MSE (11.874 and 11.827), respectively. The age of pavement, precipitation, and initial PCI are recognized as the most influential features in the sensitivity analysis. The results of the research furnish intelligent, data‐driven, and sustainable infrastructure management to enable pavement agencies to optimize maintenance planning, minimize lifecycle cost, and improve pavement resilience.

AIに質問

Bookmark

View Full Paper