Development of Roughness Prediction Models for Concrete Roads Using Machine Learning Techniques

抄録

Timely maintenance is essential for providing surface smoothness, a critical metric for evaluating pavement functional performance and road safety. This paper presents a realistic approach for estimating road roughness or longitudinal profile in terms of the International Roughness Index (IRI), specifically for concrete pavements. Nine parameters were identified under three specific domains covering pavement, traffic, and climatic characteristics, which have significant influences on roughness. Four machine learning-based models, namely Random Forest, Gradient Boosting, XG Boost, and Artificial Neural Network, were developed to predict the IRI from these parameters. The database was prepared from field measurements of in-service concrete pavements and also from the LTPP data. The results indicated that the annual average temperature and longitudinal crack length are the two most important parameters affecting the IRI. The model predictions were compared, and it was found that the Artificial Neural Network was superior to the other algorithms, followed by XG Boost.