Machine learning approach for pavement performance prediction

TLDR

Machine learning is increasingly used to predict pavement condition for optimal maintenance planning, yet most studies focus on neural networks for accuracy while overlooking other algorithms and the need for generalisable models. This paper proposes a general machine learning framework for developing pavement performance prediction models within pavement management systems. The framework employs random‑forest models trained on Long‑Term Pavement Performance data—including past IRI, structural, climatic, and traffic variables—to predict 5‑ and 10‑year IRI while prioritising generalisation across algorithms.

Abstract

In recent years, there has been an increasing interest in the application of machine learning for the prediction of pavement performance. Prediction models are used to predict the future pavement condition, helping to optimally allocate maintenance and rehabilitation funds. However, few studies have proposed a systematic approach to the development of machine learning models for pavement performance prediction. Most of the studies focus on artificial neural networks models that are trained for high accuracy, disregarding other suitable machine learning algorithms and neglecting the importance of models' generalisation capability for Pavement Engineering applications. This paper proposes a general machine learning approach for the development of pavement performance prediction models in pavement management systems (PMS). The proposed approach supports different machine learning algorithms and emphasizes generalisation performance. A case study for prediction of International Roughness Index (IRI) for 5 and 10-years, using the Long-Term Pavement Performance, is presented. The proposed models were based on a random forest algorithm, using datasets comprising previous IRI measurements, structural, climatic, and traffic data.