Abstract
The main purpose of this research is to analyze the ability of artificial neural network algorithm to estimate the unconfined compressive strength parameter of poor sand stabilized with Natural pozzolanic geopolymer. Due to the importance of sandy soil in engineering projects, this type of soil has been used. The nature of this soil is poor, first it is stabilized using geopolymer. To predict the desired parameter, the artificial neural network method was used. For the construction of the networks, 140 samples obtained from the laboratory were used. Three artificial neural networks are trained and analyzed, multilayer perceptron and cascade with Levenberg-Marquart, Bayesian regularization and gradient descent, radial basis function. After the construction and implementation of the artificial neural networks, their performance was studied and analyzed based on MSE parametric criteria and linear regression. Different networks were able to predict UCS with different accuracies. It was observed that RBF artificial neural network is more accurate in predicting this parameter. Finally, a sensitivity analysis was performed between the input parameters. Sensitivity analysis showed that treatment period plays the most important role in predicting UCS using artificial neural network.