Search

In the present work, an attempt is made to find the most suitable prediction model for Marshall Stability and the optimistic Bitumen Content (BC) in carbon fiber reinforced asphalt concrete for flexible pavements by performing Marshall Stability tests. Further the prediction analysis is performed by taking the cognizance of the published research articles. Twofold approaches are adopted; first, to find the most suitable model to predict the Marshall Stability and second to obtain the optimum binder content responsible for the highest strength. Further, to find the most suitable model for closer prediction of Marshall Stability, eighteen input parameters i.e., Binder Content (BC with fifteen variations); 4.20%, 4.30%, 4.50%, 4.90%, 5.00%, 5.10%, 5.15%, 5.20%, 5.23%, 5.30%, 5.34%, 5.40%, 5.50%, 6.00%, 6.50%, and three others i.e., Carbon fiber, Bitumen grade and Fiber length are applied in the modelling algorithm. Five Machine learning techniques viz., Support Vector Machine, Gaussian Process, Random Forest, Random Tree, and M5P model were employed to find the most suitable prediction model. Seven statistical metrices i.e., Coefficient of correlation (CC), Mean absolute error (MAE), Root mean squared error (RMSE), Relative absolute error (RAE), Root relative squared error (RRSE), Willmott's index (WI), and Nash- Sutcliffe coefficient (NSE) were used to evaluate the performance of the applied models. After performing modelling analysis, it has been found that the Random Forest-based model is outperforming amongst all applied models with CC as 0.9735, MAE as 1.1755, RMSE as 1.5046, RAE as 25.68%, RRSE as 26.93%, WI values as 0.9351, and NSE values as 0.9272 in the testing stage. The Taylor diagram of the testing dataset also conforms to the results of RF-based model. The sensitivity analysis demonstrates that binder content (BC) of about 5.0% has a significant influence on the Marshall Stability in the asphalt mix used with carbon fibers.