Prediction of optimal mixing design for stabilized soft clay soil using Artificial Neural Networks

The application of artificial simulations in predicting the behavior of materials, especially when we have real results, is very important in terms of time and cost. Therefore, in this study the data collected from unconfined compressive strength test on stabilized soil samples with lime, waste industrial and sodium silicate by neural network (GRNN) and genetic algorithm (GEP) have been investigated. Moreover, based on the results of unconfined compressive strength for the limited percentages of the experiment, simulation has been performed and verified. Then, with the development of the neural network and genetic algorithm for different states and percentages of mixing in stabilized soil, the optimized mixing percentage has been set. According to the results of genetic algorithm model, the optimal mixing design for this type of clay is 6% lime, 6% industrial waste, and 1.5% sodium silicate. The results of neural network had better predictive power than the genetic algorithm, so that the best prediction for the 90-day model of the neural network with R2 and RMSE values is 0.998 and 0.019, respectively, and the least prediction for the 7-day model of genetic algorithm with R2 and RMSE is 0.967 and 0.059, respectively.