Key points are not available for this paper at this time.
The present study assesses the proficiency of support vector machine (SVM) models utilizing four kernel functions namely normalized polynomial kernel function (SVM-NormPoly), radial basis kernel function (SVM-RBF), polynomial kernel function (SVM-poly) and Pearson universal VII kernel function (SVM-PUK), as well as artificial neural network (ANN) models in predicting the Marshall stability of polypropylene fibre reinforced asphalt concrete. A total of five statistical performance indices including coefficient of correlation (CC), mean absolute error (MAE), root mean square error (RMSE), Nash-Sutcliffe model efficiency coefficient (NSE) and scattering index (SI) were employed to statistically assess each model's performance. The statistical indicators reveal that the ANN based model demonstrates superior performance, as evidenced by their following values: CC (0.8935), MAE (1.3298), RMSE (1.8303), NSE (0.797543), and SI (0.133084), while SVM-PUK based model also demonstrated viable prediction performance over SVM-poly, SVM-RBF and SVM-NormPoly based models. Likewise, sensitivity analysis performed to investigate the significance of individual input parameter suggested that bitumen content (BC) has the utmost dominance in Marshall stability prediction while on contrary, other parameters such as polypropylene fibre length (LPPF), polypropylene fibre percentage (PPF%) and bitumen grade (BG) are least dominating parameters. From the findings of the models that have been implemented in the present study, it can be deduced that the Marshall Stability values can be effectively calculated using soft computing techniques in situations when doing so experimentally would be impractical due to the associated costs, time, or labour
Jalota et al. (Thu,) studied this question.