Prediction of pavement roughness using a hybrid gene expression programming-neural network technique

Effective prediction of pavement performance is essential for transportation agencies to appropriately strategize maintenance, rehabilitation, and reconstruction of roads. One of the primary performance indicators is the international roughness index (IRI) which represents the pavement roughness. Correlating the pavement roughness to other performance measures has been under continuous development in the past decade. However, the drawback of existing correlations is that most of them are not practical yet reliable for prediction of roughness. In this study a novel approach was developed to predict the IRI, utilizing two data sets extracted from long term pavement performance (LTPP) database. The proposed methodology included the application of a hybrid technique which combines the gene expression programming (GEP) and artificial neural network (ANN). The developed algorithm showed reasonable performance for prediction of IRI using traffic parameters and structural properties of pavement. Furthermore, estimation of present IRI from historical data was evaluated through another set of LTPP data. The second prediction model also depicted a reasonable performance power. Further extension of the proposed models including different pavement types, traffic and environmental conditions would be desirable in future studies.