A SMOTified extreme learning machine for identifying mineralization anomalies from geochemical exploration data: a case study from the Yeniugou area, Xinjiang, China

Extreme learning Machine (ELM) is a novel supervised machine learning algorithm, which has the advantages of fast-learning speed, good generalization, high classification performance, and can avoid problems such as local minimum, unreasonable learning rate, excessive number of iterations and overfitting. However, its classification performance is affected by imbalanced training data. To solve this problem, the synthetic minority oversampling technique (SMOTE) was integrated with the ELM algorithm to construct a hybrid algorithm, called SMOTified ELM, to identify polymetallic mineralization anomalies from the 1: 50,000 drainage sediment survey data in the Yeniugou area of Tokexun County, Xinjiang, China. A comparison between the SMOTified ELM model and the ELM model shows that the SMOTified ELM model is superior to the ELM model in terms of receiver operating characteristic curves (ROCs) and area under the (ROC) curves (AUCs). The ROC curve of the SMOTified ELM model is closer to the upper left corner of the ROC space than that of the ELM model. The AUC value of the SMOTfied ELM model (0.963) is higher than that of the ELM model (0.898). The polymetallic mineralization anomalies identified by the SMOTified ELM model account for 10.61% of the study area and contain 100% of known polymetallic deposits. The polymetallic mineralization anomalies identified by the ELM model account for 8.00% of the study area and contain 89% of known polymetallic deposits. Therefore, the SMOTified ELM method is a potentially useful technique for building a supervised mineralization anomaly identification model with high performance.