Comparative Study of Hybrid Artificial Intelligence Approaches for Predicting Hangingwall Stability

Abstract

Five hybrid artificial intelligence (AI) approaches based on machine learning (ML) and metaheuristic algorithms were proposed to predict open stope hangingwall (HW) stability. The ML algorithms consisted of logistic regression (LR), multilayer perceptron neural networks (MLPNN), decision tree (DT), gradient boosting machine (GBM), and support vector machine (SVM), and the firefly algorithm (FA) was used to tune their hyperparameters. The objectives are to compare different hybrid AI approaches for HW stability prediction and investigate the relative importance of its influencing variables. Performance measures were chosen to be the confusion matrix, the receiver operating characteristic (ROC) curve, and the area under the ROC curve (AUC). The results showed that the proposed hybrid AI approaches had great potential to predict HW stability and the FA was efficient in ML hyperparameters tuning. The AUC values of the optimum GBM, SVM, and LR models on the testing set were .855, .816, and .81, respectively, denoting that their performance was excellent. The optimum GBM model with the top left cutoff or the Youden’s cutoff was recommended for HW prediction in terms of the accuracy, the true positive rate and the AUC value. The relative importance of influencing variables on HW stability was obtained, in which stope design method was found to be the most significant variable.