Prediction of Pavement Performance: Application of Support Vector Regression with Different Kernels

TLDR

Pavement performance modeling is essential for pavement management, and its accuracy depends on effective variables and the chosen mathematical method. The study evaluates the capability of support vector machine (SVM) to predict future pavement condition. The authors constructed five SVM kernel models using nine extracted input variables and evaluated them against the international roughness index. The Pearson VII Universal kernel achieved accurate predictions of pavement performance throughout its life cycle.

Abstract

The pavement performance model is a basic part of the pavement management system. The prediction accuracy of the model depends on the number of effective variables and the type of mathematical method that is used for modeling the pavement performance. In this paper, the capability of the support vector machine (SVM) method is analyzed for predicting the future of the pavement condition. Five kernel types of SVM algorithm are formed and nine input variables of the proposed models are extracted from the range of effective variables on the pavement condition. The international roughness index is used as the pavement performance index. The results show that the Pearson VII Universal kernel can accurately predict pavement performance in its life cycle.