Indicator element selection and geochemical anomaly mapping using recursive feature elimination and random forest methods in the Jingdezhen region of Jiangxi Province, South China

Abstract

Determining indicator element association for mineralization can not only improve mineral exploration efficiency but also reduce the cost of unnecessary element analysis during geochemical exploration. This study provides a case study of Zhuxi tungsten-copper deposits and presents a workflow using recursive feature elimination and random forest methods to select the indicator element association for copper and tungsten mineralization in regional geochemical mapping. First, a training dataset containing positive and negative samples was built based on the known mineral deposits and mineral deposit model. Second, a 100-time simulation of recursive feature elimination with cross-validation based on random forest (RFECV-RF) was run to get a robust result of indicator elements by the ranking of variable importance. Third, the random forest (RF) method was used to integrate six indicator elements for mapping geochemical anomaly. The Youden index and prediction-area (P-A) plot were used to determine the threshold value for geochemical anomaly identification. The results demonstrated the hybrid workflow was useful to determine key indicator element for geochemical anomaly identification associated with copper and tungsten mineralization. Bi, Mo, Cu, Cd, W, and As were selected as the key indicator elements for geochemical exploration of Cu–W mineralization. Bi, Mo, W and Cu elements correspond to skarn and altered granite mineralization at depth while Cd and As elements correspond to the hydrothermal-vein mineralization at shallow levels. The result of receiver operating characteristic (ROC) curve showed that geochemical anomaly identified using the hybrid method proposed in this study had the best performance in producing comprehensive geochemical signatures. The six indicator elements also exhibited an excellent performance of identifying geochemical anomaly associated to Cu–W mineralization. This study provides a cost-benefit solution to reduce the cost of unnecessary elements concentration detection by determining a small number of key indicator elements using machine learning methods in the regional geochemical mapping for discovering mineral deposits.