Determination of Unbound Granular Material Resilient Modulus with MARS, PLSR, KNN and SVM

Estimation of reliable resilient modulus ( M-R) , a strength performance index for pavement design, is highly germane for safe design and construction of pavement. In the present study, four artificial intelligence techniques namely, multivariate adaptive regression splines (MARS), partial least squares regression (PLSR), K nearest neighbor (KNN) and support vector machine (SVM) were utilized to develop prediction models to estimate reliable M-R of unbound granular material and their perfor-mances were compared with one other. The data obtained from long-term pavement performance (LTPP) from the United States of America were used for the analysis. Results obtained from the study show that the M-R model developed with the MARS artificial intelligence technique has an explicit model when compared with the other developed artificial intelligence technique models. Furthermore, in terms of prediction accuracy, the MARS regression model, based on the results of the statistical error indices used to evaluate the developed models, gave the best prediction model. This was closely followed by the KNN model. The next was the PLSR model and the least was the SVM model. Specifically, the R-2 value of the MARS model was observed to be greater than 0.9, which indicate that the MARS model is highly reliable, and can be adequately utilized for the estimation of M-R of unbound granular material.