A comparison of prediction and classification models of unplanned stope dilution in open stope design

Unplanned stope dilution affects the cost of open stoping mining operations. The stability graph and equivalent linear overbreak slough graph have been in use for designing the stope dimensions that minimize unplanned dilution. However, despite extensive effort made by researchers, the graph methods still suffer from limitations. One of them is the poor generalization ability of these graphs. Hence, in this paper, Artificial Neural Network (ANN) based graphs are proposed as alternative tools capable of relating accurately the dilution to the stope dimensions and the rock mass quality. Two models were developed: a predictive (fitting) model and a classification model. The results indicate that ANN, as a prediction model, is not appropriate for dilution because of a low prediction capability observed (R² = 0.57). On the other hand, the classifier model showed good classification accuracy (84% on the average). In addition, the output of the classifier was used to determine the probability of unplanned dilution occurrence in the form of maps, which are useful in stope design. Therefore, the use of ANN-based classifier for open stope design is suggested while predictive models (i.e. fitting models) should be discouraged.