Prediction of Uniaxial Compressive Strength from Drilling Parameters Using Metaheuristic-Optimized Ensemble Regression Models

The uniaxial compressive strength (UCS) of rocks is a crucial parameter for assessing the quality of surrounding rock in mining, tunneling, and other geotechnical engineering applications. This study proposes a novel approach to predict UCS from drilling parameters using metaheuristic-optimized ensemble regression models. A comprehensive dataset of 345 rock samples with varying strengths was compiled through a literature review, incorporating penetration rate (ROP), revolutions per minute (RPM), torque, and thrust as inputs, with UCS as the output. Particle Swarm Optimization (PSO) was employed to optimize hyperparameters of four ensemble models: extreme gradient boosting (XGBoost), random forest (RF), backpropagation neural network (BP), and support vector regression (SVR). The dataset was split into training and testing datasets with an 80:20 ratio. Model performance was quantified by coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). Sensitivity analysis revealed that ROP and RPM were the most influential parameters for UCS prediction, while thrust showed the least impact. Results demonstrate that the stacking model achieved the optimal performance when PSO-BP, PSO-SVR, PSO-RF, and PSO-XGBoost were used as base models and linear regression served as the meta-model.