COMPARATIVE ASSESSMENT OF VARIOUS ARTIFICIAL NEURAL NETWORK TECHNIQUES FOR ESTIMATING THE SAFETY FACTOR OF ROAD EMBANKMENTS
Keywords:Artificial neural networks, Safety-factor, Road embankments, Feedforward back-propagation, Cascade forward neural networks, General regression neural network
The slope-stability analysis is one of the most important parameters for ensuring a safe design of road embankments. Currently, various traditional approaches to computing this variable can be seen in the literature. Among them, the finite element method is considered an accurate way to define the safety factor of road embankments. Previous research has investigated the capability of artificial neural networks for rapid safety-factor estimation to overcome the long process of modeling and calculations required in the aforementioned approach. However, most of these studies have focused on a single type of neural network and did not investigate the capabilities of other approaches. Therefore, this study is intended to evaluate the performance of various artificial neural network techniques in predicting the safety factor of road embankments. Within this context, the feed-forward back-propagation, cascade forward neural networks, and general regression neural network results will be compared and benchmarked against various methods used to predict this parameter. Moreover, it is intended to report the influence of neural network architecture on the accuracy of the estimation. Generally, the study results have shown that an artificial neural network provides a rapid and accurate method for calculating road embankments' safety factors. Besides, the best neural network model achieved a coefficient of determination of about 0.91 and a root mean square error of 0.236, which proves the efficiency of this technique. Moreover, the reliability assessment by comparing the neural network models against the traditional methods has shown that they provide better agreement with the finite element technique.