Two-Stage Support Vector Classifier and Recurrent Neural Network Predictor for Pavement Performance Modeling

Abstract

Accurate prediction of pavement performance is essential to a pavement infrastructure management system. The prediction process usually consists of classifying sections into families and then developing prediction curves or models for each family. Artificial intelligence, especially machine learning algorithms, provides a medium to investigate techniques that address these management concerns. This paper presents a two-stage model to classify and accurately predict the performance of a pavement infrastructure system. First, sections with similar characteristics are classified into groups using a support vector classifier (SVC). Next, a recurrent neural network (RNN) uses the classification results from the first stage in addition to other performance-related factors to predict performance. A case study using the Minnesota Department of Transportation (MnRoad) test facility database shows that the proposed model is a good classification decision support system, has better prediction results than the single-stage RNN model, and captures all underlying effects of the different variables. The significance and a sensitivity analysis of the model parameters are also presented.