Assessment of the Rock Elasticity Modulus Using Four Hybrid RF Models: A Combination of Data-Driven and Soft Techniques

The determination of the rock elasticity modulus (EM) is an indispensable key step for the design of rock engineering problems. Traditional experimental analysis can accurately measure the rock EM, but it requires manpower and material resources, and it is time consuming. The EM estimation of new rocks using former published empirical formulas is also a possibility but can be attached of high uncertainties. In this paper, four types of metaheuristic optimization algorithms (MOA), named the backtracking search optimization algorithm (BSA), multi-verse optimizer (MVO), golden eagle optimizer (GEO) and poor and rich optimization algorithm (PRO), were utilized to optimize the random forest (RF) model for predicting the rock EM. A data-driven technology was used to generate an integrated database consisting of 120 rock samples from the literature. To verify the predictive performance of the proposed models, five common machine-learning models and one empirical formula were also developed to predict the rock EM. Four popular performance indices, including the root-mean-square error (RMSE), mean absolute error (MAE), the coefficient of determination (R2) and Willmott’s index (WI), were adopted to evaluate all models. The results showed that the PRO-RF model has obtained the most satisfactory prediction accuracy. The porosity (Pn) is the most important variable for predicting the rock EM based on the sensitive analysis. This paper compares the performance of the RF models optimized by using four MOA for the rock EM prediction. It provides a good example for the subsequent application of soft techniques on the EM and other important rock parameter estimations.