g-ROKY: Maden Kaynak Kestiriminde Homojen Ortam Modellenmesine Yönelik Yeni Bir Yöntemin Geliştirilmesi

Abstract

One of the crucial steps of mineral resource estimation is to examine and determine the presence of domains which have similar grade, lithological and/or mineralogical distribution in mineral deposits. It is well known that defining the domains improves estimation quality. The methods traditionally used in modeling domains only take categorical variables of domains and their spatial relations into account. The objective of thesis is to develope new methods considering grade variable in domain modeling. For this purpose, two methods are developed, based on combination of Random Forest (RF) and Ordinary Kriging. The first method, called as ROKY (Random Forest + Kriging); includes the stages of defining the relationship between grades and categorical variable by RF, estimating the grade values at unsampled locations by Ordinary Kriging and feeding the estimated values into RF. The second method, g-ROKY (Grouped Random Forest + Kriging); is developed for avoiding the smoothing effect of kriging which causes misclassifications. g-ROKY, different from ROKY, is based on grouping the sample and kriged estimates in standardized form, constructing RO by grouped data and finally determination of the domains. A MATLAB script is written in order to apply the two methods and to analyze the results obtained. The methods are applied to public domain Jura data set and also İğdekuzu Lignite Field. In addition, the classifications made by these two methods are compared with traditional Indicator Kriging method. Within the compared methods, g-ROKY is the method that produces the closest estimation values to true domains.