A machine learning study of the dynamic modulus of asphalt concretes: An application of M5P model tree algorithm

Abstract

Dynamic modulus of asphalt concrete, which is a key parameter characterizing its performance, can be either measured in the laboratory through time-taking and expensive experiments or estimated through predictive models. In this study, the M5P model tree algorithm is used to develop the predictive models of the dynamic modulus of asphalt concretes. Compared to other machine learning-based techniques, the M5P algorithm is easy-to-use and provides more understandable linear mathematical expressions between the input and output variables. To develop the predictive models, a dataset containing information on the binder properties, gradation characteristics, volumetric properties, and test conditions was collected from the literature. Due to the highly skewed distribution of the output variable (i.e., dynamic modulus), separate predictive models were developed using the real and logarithmic values of dynamic modulus. The performance of the developed models was evaluated and compared with that of the Witczak, Hirsch, Al-Khateeb et al., and ANN models as the most commonly accepted predictive models of dynamic modulus. The findings of this study show that the models developed using the M5P algorithm outperform the previously developed models. Moreover, a logarithmic transformation of the dynamic modulus values significantly improved the performance of the model.