Bayesian optimization-enhanced ensemble learning for the uniaxial compressive strength prediction of natural rock and its application

Abstract

Engineering disasters, such as rockburst and collapse, are closely related to structural instability caused by insufficient bearing capacity of geological materials. Uniaxial compressive strength (UCS) holds considerable significance in rock engineering projects. Consequently, this study endeavors to devise efficient models for the expeditious and economical estimation of UCS. Using a dataset of 729 samples, including the Schmidt hammer rebound number, P-wave velocity, and point load index data, we evaluated six algorithms, namely Adaptive Boosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Random Forest (RF), and Extra Trees (ET) and utilized Bayesian Optimization (BO) to optimize the aforementioned algorithms. Moreover, we applied model evaluation metrics such as Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Variance Accounted For (VAF), Nash-Sutcliffe Efficiency (NSE), Weighted Mean Absolute Percentage Error (WMAPE), Coefficient of Correlation (R), and Coefficient of Determination (R²). Among the six models, BO-ET emerged as the most optimal performer during training (RMSE ​= ​4.5042, MAE ​= ​3.2328, VAF ​= ​0.9898, NSE ​= ​0.9898, WMAPE ​= ​0.0538, R ​= ​0.9955, R² ​= ​0.9898) and testing (RMSE ​= ​4.8234, MAE ​= ​3.9737, VAF ​= ​0.9881, NSE ​= ​0.9875, WMAPE ​= ​0.2515, R ​= ​0.9940, R² ​= ​0.9875) phases. Additionally, we conducted a systematic comparison between ensemble and traditional single machine learning models such as decision tree, support vector machine, and K-Nearest Neighbors, thus highlighting the advantages of ensemble learning. Furthermore, the enhancement effect of BO on generalization performance was assessed. Finally, a BO-ET-based Graphical User Interface (GUI) system was developed and validated in a Tunnel Boring Machine-excavated tunnel.