Interpreting the effects of firefly algorithm (FA) to optimize varying machine learning models for the prediction of rock joint shear strength

Abstract

Abstract To propose an effective and accurate model for the prediction of the shear strength of rock mass joint, the present study focuses on the comparison of different machine learning (ML) models, including the support vector machine (SVM), decision tree (DT), random forest (RF), K-nearest neighbors (KNN), and mixed logistic regression (MLR) models. The firefly algorithm (FA) was employed to tune the hyper-parameters of the ML algorithms, using the input parameters of the joint roughness, compressive strength, normal stress, and friction angle. The prediction performance showed that the developed model can effectively and reliably tune the hyper-parameters of the ML algorithm and arrive at the optimal structure to estimate the shear strength of the rock mass joint. Among the five ML algorithms aiming for the estimation of the shear strength, the root mean square error (RMSE) values (the training set is 0.08 and the testing set is 0.1854) of the SVM model are the lowest, and the correlation coefficient (R) values (the training set is 0.9861 and the testing set is 0.9457) are the highest, and there is no over-fitting in the prediction process. Response analysis shows that normal stress is the most influential coefficient affecting the rock mass joint shear strength, while compressive stress is the least.