Predicting the Pavement Serviceability Ratio of Flexible Pavement with Support Vector Machines

Determining pavement serviceability ratings (PSR) for flexible pavement is an important task in pavement management. The PSR is influenced by rutting depth, cracking and patching. In the existing evaluation methods, systematic analysis and regression modeling analysis are often used. However, the first method relies more on experienced judgment and the second method is restrained to use by the territory condition. In this study, a support vector machine (SVM), a new type of learning algorithm based on statistical theory, is developed to estimate the PSR of flexible pavement. The SVM uses a regression technique based on an insensitive loss function. The values predicted by SVM are compared with the values from the AASHO model and the ANN model. The study shows that SVM has better performance in estimating the PSR of flexible pavement than the other models, and it can be a useful and practical tool for estimation of PSR.