Application of data mining in pavement performance modelling: a case study

This paper compares two different pavement performance measures, one based upon ‘excessive’ maintenance costs and the second upon roughness progression deterioration rates. The purpose of this research was to utilise data mining to discover information about the factors that impact upon these performance measures. The results were compared with an existing mechanistic empirical pavement performance model, the ARRB TR Network Level Roughness Deterioration Model (ANLRD). The roughness progression model outperformed that based on excessive maintenance costs with regards to prediction accuracy. The data mining analysis also provided a series of rules that stated the most important variables upon the performance of the road section were average annual daily traffic; heavy vehicle number; and, environment, based on either the climate or soil type and rainfall. In comparison ANLRD uses combinations of the same variables, the only exception being the modified structural number(SNC), which was not available for inclusion in the data mining analysis. (a)