Estimation of copper-molybdenum anomaly values using Mahalanobis distance separation method and three commonly used data mining methods; Case study: Zafarghand


1 Department of Mining Engineering, Amirkabir University of Technology (Tehran Polytechnic)

2 Assistant Professor, Department of Mining Engineering

3 Associate Professor, Department of Mining Engineering


In order to reduce errors and save money and energy, this research has dealt anomaly separation. The importance of detecting anomaly values from the background is not hidden from anyone. For this purpose, several methods have been invented, among which we can mention Mahalanobis distance separation method, which is an effective and multi-variable method for separating anomalous values from the background. In the present study, the performance of the combination of the above separation method with three data mining methods; K-Nearest Neighbor, Simple Bayes Classifier and Convolutional Neural Network is investigated. In this way, after separating the anomalous values of copper and molybdenum in the case of 177 samples obtained from the surface sampling operation in the area of Zafarghand with the Mahalanobis distance method, in order to predict these values for each random sample, The above three data mining methods were used. The results show that K-Nearest Neighbor method is much stronger, because in the network designed by this method, no sample has been wrongly identified, which shows the high accuracy of the designed network. It should be noted that the number of wrongly predicted samples for convolutional neural network and Bayes methods are reported as 2 and 3, respectively. Considering the far more acceptable error rate for the network designed by the combination of K-Nearest Neighbor method and Mahalanobis intervals, the said combination has been introduced to the decision makers of this industry as a reliable and useful method to reach the most accurate predictions.


Main Subjects