Development of a Machine Learning-Based Tool to Predict Asphalt Concrete Overlay Performance in Hot and Humid Climates

Document Type

Article

Publication Date

11-1-2025

Abstract

Highway state agencies can achieve substantial budget savings through the strategic allocation of resources for pavement maintenance and rehabilitation activities. To achieve this, advanced prediction models are required that are capable of handling extensive field data and accurately forecast long-term pavement conditions. However, traditional pavement performance prediction models have raised concerns regarding their efficiency and accuracy, as they rely on a limited set of explanatory variables and were primarily designed for short-term prediction periods. The objective of this study was to introduce a robust decision-making tool that can predict the field performance of asphalt concrete (AC) overlays on asphalt pavements in southern states in the United States for up to 11 years. The field performance was measured in terms of the pavement condition index (PCI). This tool was designed to evaluate the predictive accuracy of multiple machine learning algorithms, including decision-tree, random forest, eXtreme gradient boosting, categorical boosting, artificial neural network, and an ensemble-learning method. Each algorithm was utilized to progressively develop six models based on historical pavement condition data from the Louisiana Department of Transportation and Development pavement management system database. These models were trained on 892 log miles of randomly selected AC overlay sections in Louisiana, with the output being the future PCI of AC overlays at a biannual rate from one to 11 years. The results indicated that the ensemble learning method achieved the highest accuracy among all algorithms, making it a viable alternative for state pavement agencies to upgrade from traditional performance prediction methods to more advanced models.

Publication Source (Journal or Book title)

Journal of Computing in Civil Engineering

This document is currently not available here.

Share

COinS