Application of deep neural networks in Geo-environmental Engineering

 In the last decade, overcoming the detrimental effects of non-recyclable materials is became a global concern. Polyethylene-Terephthalate (PET) is one of the non-recyclable materials used to produce liquid containers. The use of such materials in soil improvement has acquired importance. This study developed a tree-based predictive model for shear strength improvement caused by PET elements. To predict shear strength, a series of parametric studies led to the development of four models, i.e., DT, RF, XGB, and AdaBoost. These parametric studies aimed to determine which hyperparameters are the best for tree-based models. In spite of the fact that DT and RF are among the most powerful prediction models, XGBoost and AdaBoost offer better results. Due to their ability to learn from mistakes, they are more robust than DTs and RFs, which are semi-stochastic. According to the AdaBoost and XGBoost models, the AdaBoost model with R2train=0.99 and R2test=0.98 performed better when compared with the XGBoost model.