Enhancing Machine Learning Algorithms to Assess Rock Burst Phenomena

Abstract

One of the main challenges that deep mining faces is the occurrence of rockburst phenomena. Rockburst risk assessment with the use of machine learning is currently gaining increased attention, due to the fact that outperforms the widely used empirical approaches. However, the limited and imbalanced instance records, combined with the multiparametric nature of the phenomenon, can lead to unstable estimations. This study focuses on the enhancement of the prediction performance of five machine learning algorithms, including Decision Trees, Naïve Bayes, K-Nearest Neighbor, Random Forest and Logistic Regression, by utilizing the oversampling technique SMOTE (Synthetic Minority Oversampling TEchnique).The initial database consists of 249 rockburst incidents, from which approximately 70% was used as the training set and the remaining 30% as the test set. Parametric analyses were conducted regarding different indicator combinations, such as the maximum tangential stress, the rock's uniaxial compressive and tensile strength, the stress coefficient, two brittleness coefficients and the elastic energy index. The models were trained with the original dataset and afterwards a gradual increase of the database with synthetic instances was made until the obtainment of a balanced dataset. Subsequently the creation of synthetic instances was continued until the real incidents used for training and the synthetic incidents were of the same amount. The results from the following analysis show that SMOTE technique has a considerable effect in the evaluation metrics of the models, even after the balancing of the dataset, and can be a valuable asset for the rockburst prediction.

Evolution state prediction of step-like landslide based on an optimized intelligent model

“…The SMOTE over-sample or under-sample the data by using creating synthetic points. The SMOTE has been used to solve imbalanced rock classification in tunneling engineering (Papadopoulos and Benardos 2021). In this case, the SMOTE was used to balance the mutational points and stationary points.…”