Application of Six Metaheuristic Optimization Algorithms and Random Forest in the uniaxial compressive strength of rock prediction

The uniaxial compressive strength (UCS) is one of the most important parameters for judging the mechanical behaviour of rock mass in rock engineering design and excavation such as tunnels, sub-ways, drilling, slopes and mines stability. However, an obvious deficiency of traditional experimental operations to obtain UCS is that it suffers from a lack of efficiency and accuracy. Therefore, the prediction of the UCS of rock is of high practical significance in reducing evaluation time and improving the precision of results. At the same time, breaking the universality problem of traditional empirical models and improving the accuracy of artificial intelligence models need to absorb and accommodate more rock samples. Hence, a total of 226 rock samples with five properties were carried out from four published studies and selected to generate a dataset in this investigation, i.e., Granitic, Caliche, Schist, Sandstone and Grade III granitic. Five individual parameters of rock samples consisting of Schmidt hardness rebound number (SHR), P-wave velocity (Vp), point load strength (Is(50)), porosity (Pn), and density (D) were used to predict UCS. In this paper, six metaheuristic optimization algorithms were utilized to improve the performance of the Random Forest (RF) model, i.e., slime mould algorithm (SMA), chameleon swarm algorithm (CSA), transient search optimization (TSO), equilibrium optimizer (EO), social network search (SNS) and student psychology based optimization algorithm (SPBO). Four performance indices , the root mean square error (RMSE), the determination coefficient (R2), Willmott's index (WI) and the variance accounted for (VAF) were utilized to evaluate the performance of all models in forecasting the UCS of rock. The results of the performance comparison demonstrated that the TSO-RF model has the highest values of R2 (train: 0.9923 and test: 0.9753), WI (train: 0.9980 and test: 0.9937), and VAF (train: 99.2272% and test: 97.6852%), the lowest values of RMSE (train: 3.8313 and test: 6.5968) compared to the other models. The research in this study provided an effective attempt to further improve the accuracy of UCS prediction.(c) 2022 Elsevier B.V. All rights reserved.