Mineralized-anomaly identification based on potential-enhanced positive-unlabeled bagging: A case study from Guangxi gold ore province

Stream sediment geochemistry data has emerged as a valuable tool for identifying potential mineral deposits. However, there is a severe imbalance in the distribution of known (positive) and unknown (negative) mineralization samples in these data. To address this challenge, this study proposes a potential-enhanced positive-unlabeled bagging (PEPUB) algorithm, which takes into account the potential contribution of target mineral deposits in unsampled areas. Furthermore, the PSO algorithm is employed to optimize the model's hyperparameters and enhance its performance. To validate the proposed method, we conducted tests in the Guangxi Zhuang Autonomous Region, situated on the southwestern edge of the South China Block. The results indicate that the PEPUB-based method demonstrates superior performance in predicting potential gold deposits, with an F1 score of 0.928 and a precision rate of 90.6 %. Additionally, the obtained mineralization anomaly maps show a high level of agreement between the predicted mineralization points and the known mineralization points. This model not only facilitates the identification of potential target mineral deposits and improves exploration efficiency but also offers valuable insights into the application of semi-supervised learning in mineral exploration.