Data-Driven Estimation of Soil Internal Friction Angle Using Machine Learning Techniques

The internal friction angle of soil is considered one of the most influential parameters in geostructure design and plays a core role in assessing its safety and stability. In the current study, a precise prediction model for the internal friction angle of soil was established using high-order machine learning techniques such as Random Forest, XGBoost, and Extra Trees Regressor. The total number of ۲۴۵ soil samples taken from numerous boreholes formed a dataset, and standard penetration test (SPT) N-value, the soil type, the unit weight, the void ratio, the elevation, and the thickness of the soil cover were utilized as the input for the models. The performance outcome for the models using the statistical indices such as the coefficient of determination (R²), the Mean Absolute Error (MAE), the Mean Squared Error (MSE), and the Root Mean Squared Error (RMSE) verified that the Random Forest algorithm had the optimum performance on the test set with the outcome for R² = ۹۳.۷۵%, MAE ۱.۲۹۰۹, MSE = ۳.۱۳۴۱, and RMSE = ۱.۷۷۰۳. The above approach can be deemed a swift, cost-effective, and accurate methodology compared with the traditional methodology and can be utilized as a replacement for the traditional methodology for geotechnical professionals.

Keywords:

Machine Learning , Soil Internal Friction Angle , Data-Driven Model , Geostructure