Prediction of Undrained Shear Strength of Crushed Tire Mixture with Fine-Grained Soil by using Machine Learning Approaches

This research examines the ability of soft computing approaches (i.e. Linear Regression (LR), Gaussian Process regression (GP), Adaptive neuro-fuzzy inference system (ANFIS),Support Vector Machine (SVM) and deep neural network (DNN)) to predict the undrained shear strength (SU) of soil mixed waste crushed tires. Data set consisting of 72 different samples were used and obtained from the laboratory experiments. Out of 72 experimental observations randomly separated 50 observations were selected for model development whereas residual 22 were selected for the validation of the developed models. Input data set consist of vertical stress, percentage of the crushed tire, percentage of clay, size of clay, specific gravity of tires, Liquid limit, Plastic limit and Specific gravity of clay samples were considered as inputs whereas undrained shear strength of stabilized soil using waste crushed tires material (SU) was considered as output. Five most popular goodness fit assessment parameters were used for the comparison among developed models. Results suggest that DNN based model works superior to other developed models for the prediction of SU the soil samples mixed with tires waste material with coefficient of correlation values as 0.9975, 0.9736, Root mean square error values as 2.4198, 7.5319, Mean absolute error values as 1.8407, 6.1870, Scatter index values as 0.0311, 0.0959 and Nash Sutcliffe model efficiency values as 0.9943, 0.9387 for training and testing stage respectively. Sensitivity analyses offer that specific gravity of tires, size of clay and vertical stress were the most influencing variables in the prediction SU the soil samples mixed with tires waste material.