Estimation of Uniaxial Strength of Rock: A Comparison between Bayesian-Optimized Machine Learning Models

The uniaxial compressive strength (UCS) is a critical parameter for classifying rock types. Several laboratory methods are available to determine the UCS of rocks; however, these methods are often time-consuming, expensive, and labor-intensive. To address these limitations, this research introduces an advanced soft computing model for estimating rock UCS by comparing various approaches, including multilinear regression (MLR), Bayesian-optimized support vector machine (SVM), Gaussian process regression (GPR), decision tree (DT), and ensemble tree (ET) models. The study highlights that multicollinearity among parameters such as area, mass, density, Young's modulus, and P-wave velocity significantly influences UCS estimation. The results demonstrate that the GPR model emerges as a robust soft computing technique, achieving a performance index of 1.62, variance accounted for of 84.06%, a performance score exceeding 0.91, a mean absolute percentage error of 0.19%, and a root mean square error of 27.17 MPa. These metrics indicate superior performance compared to the SVM, DT, ET, and MLR models. Furthermore, uncertainty analysis (ranking first), error characteristic curve analysis (3.80E-03 in testing), score analysis (total score of 128), convergence analysis, accuracy metrics, reliability analysis (using A20 index, scatter index, and agreement index), and overfitting evaluation (0.9316) confirm the GPR model's dominance in estimating rock UCS. The GPR model also estimated UCS with a confidence interval of ± 75 MPa, outperforming hybrid models reported in the literature. This research offers significant value to geotechnical and rock engineers, enabling them to estimate UCS using fundamental rock properties without the need for extensive laboratory testing.