A comparative study between support vector regression and relevance vector regression for estimating the ore grade in a rare metal deposit, DPR Korea

Abstract A critical step for mineral resource exploration and evaluation is estimation of ore grade. Many traditional ore grade estimators except for artificial intelligence (AI) methods only utilize location information of measured value as the input to predict without consideration for other factors related to target, which result in decreasing prediction accuracy. This paper uses a support vector regression (SVR) and relevance vector regression (RVR) to predict the ore grade in a rare metal deposit. The main goal of this study is to examine the applicability and capability of SVR and RVR methods for estimating ore grade and to find a way to predict the ore grade of rare metals as accurately as possible, considering its unique characteristics such as symbiotic relationship between rare metal components, as well as location information. Results illustrate both methods of SVR and RVR could successfully predict the grade of target metal using the location information and grades of non-target metals associated with target one. Furthermore, RVR shows reliability more than SVR with considerable estimating and generalizing capability.