Journal of Advances in Civil and Mechanical Engineering

Modelling and Prediction of Soil Shear Strength Using Machine Learning

Abstract

Tony Ndungu Munene

This study explores the application of machine learning algorithms to predict soil shear strength parameters. Using a dataset of 188 soil samples containing properties such as Atterberg limits, grain size distribution, maximum dry density, and optimum moisture content, eight different machine learning models were trained and evaluated. The models included Linear Regression, Random Forest (baseline and tuned), XGBoost (baseline and tuned), Neural Networks, LightGBM, and Support Vector Regression. Performance was assessed using Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and R² metrics. Support Vector Regression achieved the best performance with RMSE of 0.053, MAE of 0.0262, and R² of 0.41. However, all models showed limited ability to explain variance in shear strength, with R² values below 0.5, indicating challenges in predicting this complex geotechnical parameter. The results demonstrate that while machine learning shows promise as an alternative to traditional testing methods, data quality and quantity remain critical limitations for practical implementation.

PDF