Research on prediction method of rock uniaxial compressive strength based on interpretable INFO-Stacking model
Jie Wang, Xinqiu Fang, Duanwei He, Yang Song, Ningning Chen, Haotian Feng
Abstract
Uniaxial compressive strength (UCS) is essential for evaluating rock properties and ensuring stability in geotechnical and mining engineering, frequently applied in slope design, tunnel reinforcement, and mining operations. To overcome the limitations of traditional UCS testing, including high costs and lengthy procedures, this study presents an interpretable model grounded in the INFO-Stacking algorithm. The model, trained on 285 rock samples, incorporates variables such as porosity, Schmidt rebound hardness, P-wave velocity, and point load index, with UCS as the target output. We evaluated six machine learning models, both before and after INFO optimization, and developed a stacking ensemble model combining RF, XGBoost, CatBoost, and AdaBoost. The model demonstrated superior performance compared to traditional and individual models. SHAP(Shapley Additive Explanations) analysis clarified the significance of porosity and P-wave velocity, enhancing the model's interpretability. Additionally, a user-friendly graphical interface was created to facilitate UCS predictions in real-world engineering scenarios. The INFO-Stacking model achieved an R² of 0.952, RMSE of 9.975, and MAE of 7.349, providing a more accurate and efficient approach to UCS prediction and supporting engineering decision-making.