LS-SVR method of ore grade estimation in Solwara 1 region with missing data

Seafloor hydrothermal sulphide is a new poly metallic ore resource as well as oceanic poly metallic nodules and rich-Co crust containing huge amounts of metals and rare metals. The appropriate and accurate estimation of ore grade plays an important role for the prediction of total mineral resources and further exploitation. Kriging and neural network methods have already been successfully used for grade estimation problem. However, the performance of these methods is not perfect for the limited borehole data. Therefore, this paper introduces a new nonlinear method to the issue of ore grade estimation based on the least squares support vector regression (LS-SVR). The borehole data obtained from Solwara 1 region are heterogeneous and discontinuities with huge missing values. Data preprocessing methods including weighted K-nearest neighbor (WKNN) imputation and Genetic algorithm (GA) for data segmentation were used before using LS-SVR algorithm. Finally the performance and efficacy of the LS-SVR were compared with BP, RBF neural network and geostatistical techniques such as inverse distance weight (IDW) and ordinary kriging (OK). The outcome indicates that the LS-SVR method outperforms other four methods.