A comparative study of hybrid adaptive neuro-fuzzy inference systems to predict the unconfined compressive strength of rocks

Abstract The accurate prediction of unconfined compressive strength (UCS) in rock samples is critical for the successful planning, design, and implementation of mining and civil engineering projects. UCS is crucial in assessing the stability and durability of rock masses, which directly influences the safety, efficiency, and cost-effectiveness of construction and excavation operations. Here’s a refined version of your text for enhanced clarity and flow: in this part, the execution of the proposed model was compared for both single and hybrid configurations. Hybrid models included support vector regression (SVR) combined with the Seahorse Optimizer (SVSH) and SVR combined with the COOT optimization algorithm (SVCO). For training, 70% of the UCS dataset was utilized, while the remaining 30% was equally divided between testing (15%) and validation (15%). For the model evaluation, several metrics were considered in this work, including the R 2 , RMSE, WAPE, MAE, and RAE, which ensure fairness in the analysis. The closer the R 2 value comes to 1, the better the performance. The error metrics should be close to 0 for better accuracy. From Table 2, one can observe that the result of the standalone SVR model gave an RMSE of 6.213 during training and 9.454 during testing, hence showing poor performance. However, the inclusion of optimization algorithms significantly improved the performance of the SVR framework. Among the hybrid models, the SVSH model had the best performance, with an R 2 value of 0.998 and an RMSE of 1.261 during training. The SVCO model performed moderately, with an R 2 value of 0.988 during training.