Development of a Support Vector Machine (SVM) and a Classification and Regression Tree (CART) to Predict the Shear Strength of Sand-Rubber Mixtures

Recycling waste materials such as waste tires in geotechnical projects can greatly contribute to environmental issues. An important feature of sand-rubber mixtures is their shear strength, which depends on many factors such as the size distribution of sand and rubber, density of mixtures etc. Due to the multiplicity of these effective factors, this paper evaluated the performance of two Artificial intelligence (AI) methods, namely a support vector machine (SVM) and a classification and regression tree (CART) algorithm, to predict the shear strength of sand-rubber mixtures. For this purpose, a database with 101 datasets including nine inputs and one output, i.e., the ratio of shear strength to normal stress, was used. The inputs parameters included dry density, mean particle size (D₅₀), coefficient of curvature (C_c) and uniformity coefficient (C_u) of sand, normal stress, rubber percentage, and D₅₀, C_c and C_u of rubber. The results of the best SVM and CART models were also compared with the result of multiple linear regression (MLR) method. The results show that R² for the test database was 0.90, 0.90 and 0.55 for the CART, SVM and MLR models, respectively. In addition, the MAE of CART, SVM and MLR methods were 0.013, 0.013 and 0.041, respectively. Therefore, according to the results, both AI methods have a great performance to predict the shear strength of sand-rubber mixtures.

Keywords: Sand-rubber mixtures, Shear strength, Support vector machine, Classification and regression trees.