Stability Assessment of Open Spans in Underground Entry‑Type Excavations by Focusing on Data Mining Methods - Mining, Metallurgy & Exploration (2024)

Entry-type mining methods, such as cut and fill method, play a significant role in underground mining, and they are still suitable methods for extracting deposits with irregular shapes and with weak host rocks. In these mining methods, the stability of the open spans is important for improving worker safety and increasing ore recovery. This study focuses on the application of six data mining classification methods, namely multinomial logistic regression (MLR), classification and regression trees (CART), logistic model tree (LMT), decision tree (J48), multivariate adaptive regression spline (MARS), and gene expression programming (GEP), in order to develop predictive models for stability assessment of underground entry-type excavations. For this purpose, a database containing 399 case histories of entry-type excavations from seven Canada underground mines was used. These models predict the stability status of entry-type excavations based on two input parameters, dimensions of the span and rock mass rating (RMR). The performance of developed models was evaluated using six performance metrics namely accuracy (ACC), sensitivity (SE), precision (PR), Matthew’s correlation coefficient (MCC), area under the receiver operating characteristic (ROC) curve (AUC), and performance index (PI). The results showed that all developed models can be used to predict the excavation stability, but the comparison of models indicated that the J48 model provides the highest performance with ACC = 0.900, SE = 0.869, PR = 0.922, MCC = 0.835, AUC = 0.891, and PI = 4.417. Thus, the proposed J48 model can be applied to identify the susceptibility of span instability in the entry-type mining, and if instability has been predicted, the necessary measures, including the use of suitable support systems, can be considered to reduce the casualties resulting from span instability.