Keywords: prediction, genetic algorithm, support vector machines, asphaltic concrete, Marshall stability.

The objective of the study, presented in this paper, is to explore the applicability of a support vector machine approach with different kernel functions for predicting the stability of asphaltic concrete mixes. The results of this approach are compared with back propagation and cascade correlation neural network models. All methods used the following classes of input data, which can be easily measured in the laboratory: the percentage of course aggregate, which is basalt type (igneous rock), percentage of fine aggregate, (natural sand), percentage of filler, (ordinary Portland cement type 1), and the percentage of optimum bitumen content. A genetic algorithm is used to determine optimal values of the free support vector machine parameters for different kernel functions. Among the models, the excellent performance of the support vector machine with a radial-basis-kernel-based model demonstrated the potential to function as a useful tool for the estimation of the stability of asphaltic concrete mixes to assess the maximum obtainable prediction accuracy. In conclusion, the support vector machine (radial basis function kernel) model has the highest accuracy and better generalization performance than the cascade correlation neural network and back propagation neural network models. The results obtained in this investigation demonstrate that the support vector machine (radial basis function kernel) model is a promising alternative to neural networks for the stability of asphaltic concrete mixes forecasting.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £45 +P&P)