Predicting volumetric properties and mechanical characteristics of asphalt concretes for thin wearing layers using a categorical boosting model

In recent years, new modeling strategies based on data-driven approaches are gaining increasing popularity in the field of pavement engineering. This study is aimed at developing a novel predictive model based on a supervised categorical boosting (CatBoost) algorithm that allows volumetric properties and mechanical characteristics of asphalt concretes (ACs) for thin wearing layers to be simultaneously predicted. The research involved 92 AC specimens produced both in laboratory and in plant with two different types of bitumen: a conventional and a modified one. In particular, air voids content, voids in the mineral aggregate, and stiffness modulus at 20C were successfully correlated to bitumen content, particle size parameters and a categorical variable distinguishing the mixture production site and the binder type. The best model hyperparameters were accurately determined, and several performance metrics were evaluated to confirm the remarkable predictive capabilities achieved by the developed machine learning model.