Optimal Clustering of Pavement Segments Using K-Prototype Algorithm in a High-Dimensional Mixed Feature Space

Abstract

The efficiency of pavement lifecycle planning highly depends on the accuracy of condition predictions. Therefore, transportation agencies strive to maximize the impact of the limited budget through investment decisions empowered by accurate deterioration modeling. For this purpose, family deterioration models were developed based on clustering techniques to overcome the limitations of data availability. However, most of the existing pavement clustering approaches rely on the subjective opinion of experts not only on the selection of factors contributing to deterioration but also in classifying the selected factors. Also, the impact of clustering algorithms and their configurations on the accuracy of deterioration models were marginally investigated in previous studies. To this end, we developed a clustering method and incorporated a wide mixture of categorical and continuous contributors. Then, we created a process to find the optimal configuration of clusters. Finally, we implemented the devised methodology on a large-scale case study. The comparison of our results with past studies revealed an improvement in the accuracy of the condition predictions. Consequently, this study provides a tool for accurately predicting the maintenance needs of pavements and improves the efficiency of lifecycle planning.