Unconfined Compressive Strength Prediction of Rocks Using a Novel Hybrid Machine Learning Algorithm

This paper introduces a novel methodology for predicting Unconfined Compressive Strength (UCS) in rocks by integrating Support Vector Regression (SVR) with two cutting-edge optimization algorithms: the Seahorse Optimizer (SO) and the COOT Optimization Algorithm (COOT). Unlike traditional UCS prediction methods that often struggle with slow convergence and local minima entrapment, this approach leverages the nonlinear modeling capabilities of SVR and enhances its performance through advanced optimizers. The SVSH (SVR+SO) and SVCO (SVR+COOT) models were developed and evaluated using a comprehensive dataset of rock samples with UCS measurements. Comparative analysis demonstrates that the proposed models not only achieve significantly higher prediction accuracy but also exhibit faster convergence compared to standalone SVR. These results underscore the potential of the hybrid SVR-optimizer models to set a new benchmark in UCS prediction, offering greater precision and computational efficiency. Among them, the SVSH models had the maximum accuracy with an excellent R2 value of 0.998 and a scanty RMSE value of 1.261. Therefore, the results confirm that the proposed SVSO model is a promising tool to be used by engineering and geological professionals. This ensures a very strong and reliable UCS prediction method that is vital for the improvement of civil engineering project safety and efficiency.