Pavement Condition Prediction Using Clusterwise Regression

Abstract

In this paper, the clusterwise regression model, which fits data to more than one curve, was introduced to model the deterioration of pavement condition. To make the model solvable in practice, a modification was made to the model by the membership of a pavement to each cluster being estimated with the fuzzy sets concept with the corresponding errors. The number of unknowns in the modified model is reduced significantly. The model was then extended to be applicable to cases in which more than two clusters or nonlinear equations were used. Based on the result of the modified clusterwise regression model, a procedure was proposed to predict the pavement condition rating (PCR) for any individual pavement, given that the pavement condition rating at the present age was known. In the example, the ordinary least squares regression method was first employed to determine the PCR prediction curve for a pavement group. The PCRs of the individual pavements were predicted by an adjusted prediction curve based on the prediction curve of the group. The proposed prediction procedure was then applied to the same data set to make the predictions. The results showed that the proposed procedures using the modified clusterwise regression method could result in a smaller prediction error and thus produce a more accurate prediction than one produced with the adjusted curve. Therefore, to improve the accuracy of the pavement condition predictions, the modified clusterwise regression model is recommended in cases in which the pavement families are not well defined.