Estimation of shear strength parameter of silty sand from SPT-N60 using machine learning models

ABSTRACT This study represents the angle of internal friction ( ) estimation of silty sand (SM) of Bangladesh using SPT-N, the depth of sample collection, and the grain size analysis results using machine learning models. To develop the predictive model, Multiple Linear Regression (MLR), Support Vector Regression (SVR) and Artificial Neural Network (ANN) algorithms are used. Soil samples have been collected from 210 boreholes beside the rail track of the Joydevpur-Mymensingh-Jamalpur section. The performance of the models is evaluated using the R2 score, Root Mean Squared Error (RMSE) and Mean Squared Error (MAE). According to the evaluation metrics, SVR with Radial Basis Function (Rbf) kernel performs better than ANN and MLR, and a web application is prepared providing estimated ϕ based on the user input. Later SVR is compared with the established empirical equations and shows that Wolff’s model is under-predicting and Nitish Puri’s model is over-predicting than actual ϕ. However, the model proposed in this study produces lower residual internal friction angle and improved R2 score, RMSE and MAE which can be used to predict the internal friction angle of silty sand in Bangladesh with higher precision.