Predicting Pavement Condition Index Using Machine Learning Algorithms and Conventional Techniques

Abstract

Government agencies and transportation engineers use pavement management systems (PMS) to evaluate pavement performance and keep pavement above the minimum acceptable performance standards. The Pavement Condition Index (PCI) and the international roughness index (IRI) are among the most commonly used indices to evaluate pavement conditions. Due to IRI data collection being more accessible and less expensive than collecting pavement distress data, this study aims to develop PCI models that can successfully estimate the PCI values based on IRI for flexible pavement using two Machine Learning techniques (ML), namely: Random Forest (RF), and Support Vector Machine (SVM), and three conventional techniques, namely: linear, quadratic, and cubic regression. The study was carried out with the database collected from the Long-Term Pavement Performance (LTPP) program. The results of the dataset reveal that both ML models (RF and SVM) have strong prediction ability with high values of coefficient of determination (R^2 = 99.7 and 96.8) %, and low values of Root Mean Squared Error (RMSE = 1.095 and 3.569) % and Mean Absolute Error (MAE = 0.474 and 2.244). In conclusion, the goodness of fit of the proposed ML models was compared with conventional techniques models previously developed. The results showed that the ML models yielded higher prediction accuracy than conventional techniques.