Comparison of artificial neural networks and support vector machine classifiers for 3D modeling of mineralization zones (Case study: Miduk copper Deposit)

Due to the relation of mineralization zones with grade variability in porphyry copper deposits, the preparation of the three-dimensional model of these zones is one of the pre-estimation steps in evaluation this type of deposits. The quality of this model has a significant impact on the quality of the grade estimates, the proper design of long-term extraction and ultimately reducing the problems between the mine and the processing plant. The usual way to prepare this model is to use a constrained modeling technique, which is a complex and time consuming process. One of the possible solutions for the preparation of these models is the use of unconstrained methods, such as intelligent methods. This paper attempts to study the performance of artificial neural network and support vector machine in the separation of mineralization zones (including leached, hypogene and supergene zones) in Miduk copper deposit. The northing co-ordinate, easting co-ordinate and height of the samples are used as input variables, and the observed mineralization zones in them are used as the output variable. Investigating the results of these intelligent algorithms in the separation of geological zones shows that the support vector machine classifier has a better performance than the artificial neural network. The better performance of the support vector machine method is shown by 1) the higher accuracy of this method in the training and testing stages and 2) the comparison between the block model with the grade control observations.