Rock Mechanics
Barkat Ullah; Raja Khurram Mahmood Khan
Abstract
Uniaxial compressive strength (UCS) is an essential feature for characterizing and classifying rock masses, forming a critical component of rock failure criteria with extensive applications in mining and geotechnical engineering. This study aims to evaluate the performance of different machine learning ...
Read More
Uniaxial compressive strength (UCS) is an essential feature for characterizing and classifying rock masses, forming a critical component of rock failure criteria with extensive applications in mining and geotechnical engineering. This study aims to evaluate the performance of different machine learning (ML) models in forecasting the UCS of sandstone obtained from the Murree and Kamlial formations in the Muzaffarabad area, northwestern Himalayas, Pakistan. The ML models—namely artificial neural network (ANN), adaptive neuro-fuzzy inference system (ANFIS), support vector regressor (SVR), random forest (RF), and extreme gradient boosting (XGBoost)—were developed to predict UCS (MPa) based on porosity (η), point load index (Is(50)), Schmidt hammer rebound value (Rn), and aggregate impact value (AIV) as input variables. A dataset containing 80 points was divided using a 70:30 split ratio for training and testing sets. K-fold cross-validation (with 5 to 10 folds) was employed to enhance the models' generalization ability. The performance of the models was evaluated using mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and coefficient of determination (R²). Results revealed that the XGBoost model outperformed the other models, achieving a high R² value of 0.99 and low error values for MAE (0.789), MSE (1.168), and RMSE (1.080). The overall accuracy of the models can be ranked as follows: XGBoost > RF > ANN > ANFIS > SVR. This study provides a benchmark for predicting the UCS of sandstones and similar rocks where complex geology complicates the collection of intact samples.
