Abstract: The prediction model which integrates non-destructive testing and machine learning has emerged as a hotspot for predicting tensile strength of rock. This paper presents a data-driven modeling approach for predicting rock’s tensile strength based on Multivariate Adaptive Regression Splines (MARS). An experimental dataset comprising 80 data sets, including Schmidt hammer rebound number, dry density, point load strength index, and Brazilian tensile strength, was collected. The dataset was randomly divided into training (70%) and testing (30%) sets. Additionally, artificial neural network (ANN), support vector machine (SVM), and decision tree (DT) techniques were utilized to develop machine learning models. The performance of the MARS model was compared with those of the DT, ANN, and SVM models. The predictive accuracy of the developed models was evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MSE), coefficient of correlation, and coefficient of determination. The results manifested the satisfactory predictability of the machine learning models, with the MARS model exhibiting optimal performance, followed by SVM model, ANN model, and DT model in descending order. Notably, MARS was able to assess the relative importance of each variable through ANOVA decomposition. The model presented in this paper is conducive to rapidly obtaining the tensile strength of rock.

Key words: rock mechanics, tensile strength of rock, Multivariate Adaptive Regression Splines(MARS), machine learning, prediction model