Machine-Learning-Based Framework for Prediction of the Long-Term Field Performance of Asphalt Concrete Overlays in a Hot and Humid Climate

Abstract

Pavement performance prediction models are used by state agencies to determine pavement maintenance and rehabilitation strategies. However, most performance prediction models are based on a limited number of parameters and a maximum prediction period of five years. With the ever-increasing amount of available pavement performance data, machine-learning techniques have become a promising alternative to traditional performance prediction models. The objective of this study was to develop a machine-learning-based framework for states with a hot and humid climate that can predict the long-term field performance (up to 11 years) of asphalt concrete (AC) overlays on asphalt pavements based on key project conditions. The pavement condition index (PCI) was used as the pavement performance indicator. Two machine-learning algorithms, namely, random forest (RF) and CatBoost, were examined. A total of 892 log-miles of AC overlay data were obtained from the Louisiana Department of Transportation and Development Pavement Management System database. Based on the collected data, six models were trained (for each algorithm) and validated to predict the PCI of AC overlays for up to 11 years. The results indicated that the RF algorithm yielded higher accuracy than the CatBoost algorithm. Therefore, the RF-based models were considered in the proposed decision-making framework.