Rockburst prediction in kimberlite using decision tree with incomplete data

Abstract

Abstract A rockburst is a common engineering geological hazard. In order to predict rockburst potential in kimberlite at an underground diamond mine, a decision tree method was employed. Based on two fundamental premises of rockburst occurrence, σ θ , σ c , σ t , W E T are determined as indicators of rockburst, which are also partition attributes of the decision tree. 132 training samples (with 24 incomplete samples) were obtained from real rockburst cases from all over the world to build the decision tree. The decision tree based on 108 complete samples was built with an accuracy of 73% for 15 validation samples while another decision tree based on 132 samples (with 24 groups of incomplete data) shows an accuracy of 93% for validation samples. Hence, the second decision tree was employed for kimberlite burst prediction. 12 samples from lab tests and a numerical model were used as test samples. The results indicate a moderate burst liability which matches real situations at the diamond mind in question.