Applying optimized relevance vector regression approach for indirect forecasting rock mass deformation modulus

Rock mass deformation modulus is a significant design parameter in surface/underground projects such as foundations, slopes and tunnels. Due to the presence of discontinuities in the rock mass, determining this deformation modulus by analyzing cylindrical core samples is very difficult. Since it is difficult to determine the deformability of rock masses at the laboratory scale, several in situ test techniques have been developed, such as dilatometer, plate-loading tests, and so on. However, determining the deformation modulus from in situ test techniques is highly costly, time-consuming and sometimes impossible to perform. To address this issue, in this paper, the results of the application of relevance vector regression (RVR) optimized by harmony search (HS) to create an estimation model for the indirect predicting rock mass deformation modulus was presented. The RVR-HS model was applied to data from open access literature. In this model, the deformation modulus was used as the output, while elastic modulus of intact rock (E_i), uniaxial compressive strength of intact rock (UCS), rock mass rating (RMR) and depth were as the input variables. The RVR-HS model, with MSE = 0.0172 and R² = 0.8696, is a dependable technique for forecasting the deformation modulus with a high level of precision and reliability.