Estimation of the dynamic modulus of asphalt concretes using random forests algorithm

Abstract

Dynamic modulus (|E|) of asphalt can be estimated using predictive models to avoid the time-taking and costly laboratory-based measurements. Several predictive models such as the Witczak model have been widely used by many researchers for the prediction of the |E|. Previously developed models have been widely reported to either overpredict or underpredict the values of |E|. In this study, to overcome the issues related to the previously developed models, a random forests algorithm was used to develop predictive models of the |E| using a comprehensive dataset. The performance of the developed models was compared with that of the Witczak model using an independent dataset. The results show that random forests algorithm can be successfully used to develop a model for the estimation of the |E| with better performance than of the Witczak model. The R2 values of the developed model in this study and the Witczak model were obtained as 0.9462 and 0.7371, respectively. Through a logarithmic transformation, the R2 value increased from 0.9462 to 0.9634. A sensitivity analysis was also performed to find the most significant factors that affect the |E|. The variables defined for test temperature and loading frequency were found to have the most effective impact on the prediction of |E|.