Prediction of International Roughness Index Using CatBooster and Shap Values

International Roughness Index (IRI) is the performance index of pavements that exhibits the efficiency of pavement smoothness. Road roughness is a fundamental element used for determining the performance of pavements and the ride quality of road users, therefore, this research aims to develop the precise IRI prediction model for flexible pavements using advanced machine learning algorithms including supervised methods. This research is directed toward accessing the functional performance of the pavements through long-term pavement performance (LTPP) databases. For developing the model, the incorporated dataset includes a set of functional attributes from general pavement studies (GPS-1, GPS-2 and GPS-6) and specific pavement studies (SPS-1, SPS-3 and SPS-5). The developed algorithms showed that the machine learning algorithms are more precise and accurate in predicting the IRI than the traditional regression approaches. The machine learning algorithms use the shapely additive explanation (SHAP) values to access the feature significance for each independent element on the predictive performance. The analysis showed that CatBooster Regression outperformed the random forest regression, artificial neural network (ANN), and the simple regression models in terms of mean square error and prediction quality with a coefficient of determination up to 0.99. The study depicted that it is possible to correlate the roughness index with pavement and structural, climatic and distress parameters that can be utilized for pavement maintenance.