Probabilistic Analysis of Pavement Distress Ratings with the Clusterwise Regression Method

Abstract

A mixture clusterwise regression method is proposed to analyze pavement distress ratings and predict the future performance of pavements. Rather than using a single equation as the ordinary least-squares regression method, a clusterwise regression uses several regression equations (clusters) to fit a data set with a large variation. Each cluster indicates a portion or a percentage of a data set that follows a uniform tendency. A weighted regression function consisting of all clusters is introduced to take into account current and historical observations of pavement distresses and to predict future pavement conditions. An example using the existing database is provided to compare the proposed model with the conventional Markov probabilistic model. The results show that at the network level the proposed model provides much more accurate predictions than the conventional Markov model does; at the project level, the proposed model correctly predicted five of seven ratings for an example pavement, but the conventional Markov model predicted none of them. This model, if integrated into a pavement management system, can significantly improve the accuracy and applicability of the system, especially for a complex highway system.