Generalised Clusterwise Regression for Simultaneous Estimation of Optimal Pavement Clusters and Performance Models
This paper focuses on clusterwise regression (CR) approach for modelling of pavement performance. CR simultaneously clusters the data and estimates the associated models. Previous studies using CR approach have a few limitations: (1) the explanatory power of variables used in the analyses was not tested; (2) the approach could not find the optimal number of clusters; (3) the objective function was to minimise the sum of squared errors, which is not the best to seek for the optimal number of clusters; (4) the model functional form was restricted to be either linear or nonlinear. To address these limitations, this paper proposes a generalised mathematical programme and solution algorithm within the CR framework. Bayesian Information Criteria was used as the objective function. The proposed approach explored all possible combinations of potential significant explanatory variables to select the best model specification. The potential multicollinearity issues in the models were addressed if required. Both linear and nonlinear functional forms were estimated using a large dataset in Nevada. Predictive accuracy of the resultant models was evaluated using root-mean-square error (RMSE), normalised RMSE, and mean absolute errors. The results showed that the nonlinear models were more accurate than the linear models in estimating present serviceability index.