Application of regression trees to LTPP data analysis

This study is aimed at better understanding the factors that affect asphalt pavement roughness. Using the data collected under the General Pavement Studies of the Long-Term Pavement Performance Program (LTPP), the relationship between asphalt pavement roughness and design features, site conditions, and construction factors were investigated. Two hundred and thirty-eight GPS-1 experiment sections were reviewed for the study. A methodology for developing pavement roughness prediction models for LTPP data is illustrated. The classification and regression tree (CART) analysis, a nonparametric method was used for exploring the relationship between pavement roughness and design features. The proposed methodology is a tree structure of easy interpretation and application. Significant factors are selected using tree methods, and priorities of the factors affecting pavement roughness are provided. For a specific climatic region the best design and construction practice that lead to the smallest average IRI is indicated. The results obtained are consistent with previous studies and show that the statistical approach used provides simple, accurate and reasonable results that are more applicable and practical for providing guidance in pavement design and construction.