Evaluation of Soft Computing Methods for Estimating Tangential Young Modulus of Intact Rock Based on Statistical Performance Indices

Abstract

The tangential Young modulus (E-ti) of intact rock is a critical parameter in engineering geological design calculations and rock mass classification systems. The E-ti of various rock types has been successfully estimated by many studies based on numerous soft computing methods in recent years. However, these studies mainly involve a single analysis method or are valid for a limited number of samples. For this reason, this study aimed to compare artificial neural networks (ANN), adaptive neural fuzzy inference system (ANFIS), and Gene expression programming (GEP) methods to estimate the E-ti of various rock types based on 147 datasets collected from the published literature. As a result of the soft computing analyses, three different predictive models were proposed in this study. In the proposed prediction models, rock properties such as dry density (rho(d)), effective porosity (n(e)), P-wave velocity (V-p), and uniaxial compressive strength (UCS) were used. The estimation performance of the proposed models was examined through several performance indices such as coefficient of determination (R-2), root mean square error (RMSE), the variance accounted for (VAF), and mean absolute percent error (MAPE). As a result of statistical analyses, it was determined that the ANFIS model presents a better prediction performance (R-2 = 0.967) than the other methods in the training datasets. On the other hand, the accuracy of the ANFIS model decreased significantly in the test datasets (R-2 = 0.803). Furthermore, the GEP model presented the lowest predictive performance. Finally, considering the overall estimation accuracy of the proposed models, it was concluded that the proposed ANN model with an R-2 of 0.94 could reliably be used to estimate the E-ti of investigated rocks.