Data-driven machine learning approaches for predicting the shear strength of rock joints

Accurate prediction of the shear strength of rock joints is crucial for assessing the stability of civil and mining engineering projects. Traditional methods for determining the shear strength of rock joints are time-consuming, costly, and computationally complex. Machine learning methods, which are driven by data, provide a cost-effective and rapid approach to predicting rock joint shear strength, overcoming the limitations of traditional techniques. This study employs nine machine learning models: eXtreme gradient boosting (XGBoost), random forest (RF), support vector machine (SVM), decision tree (DT), Gaussian process regression (GPR), K-nearest neighbors (KNN), categorical boosting (CatBoost), extreme learning machine (ELM), and adaptive boosting (AdaBoost). A dataset of 295 data points was compiled from an extensive set of literature. Five input features, namely, normal stress, uniaxial compressive strength, Young's modulus, joint roughness coefficient (JRC), and specimen length, were selected, with shear strength of the rock joints as the output variable. The performance of the nine ML models was assessed using the root mean square error (RMSE), coefficient of determination (R 2 ), and mean absolute error (MAE). Due to its unique ordered boosting mechanism and symmetric tree structure, CatBoost outperformed the other models, achieving RMSE, R 2 , and MAE values of 0.4663, 0.9765, and 0.3508, respectively. Compared with the experimental results, the model yielded a mean square error (MSE) of 0.0360. The proposed ML method offers a cost-effective and efficient solution for predicting rock joint shear strength. • Nine machine learning models are evaluated for predicting rock joint shear strength, with CatBoost achieving the best performance. • A dataset of 295 samples, with five key input parameters, is used for model training and testing. • CatBoost achieves superior accuracy (RMSE: 0.4663, R 2 : 0.9765) and minimal error (9.0%) in shear strength prediction.