A SMOTified-GAN-augmented bagging ensemble model of extreme learning machines for detecting geochemical anomalies associated with mineralization

Abstract

The use of supervised machine learning and deep learning techniques to automatically detect geochemical anomalies associated with mineralization has become a current research hotspot. However, due to the scarcity of known mineral deposits in the study area, the establishment of supervised machine learning and deep learning models faces the challenge of highly imbalanced data classification. To address this challenge, the SMOTified-GAN oversampling technique and bagging strategy were combined with extreme learning machines (ELMs) to construct a robust high-performance ensemble classification model for detecting geochemical anomalies associated with mineralization. In this ensemble model, SMOTified-GAN is used to balance the ratio of positive (mineralized) to negative (background) samples in geochemical exploration data set, while keeping the sample distribution pattern of positive samples unchanged. The bagging strategy is used to construct a robust ensemble model from the simple ELM classifiers to improve the robustness of the supervised anomaly detection model. Taking the Helong area (Jilin Province, China) as the case study area, three bagging ensemble models of the simple ELM classifiers with SMOTified-GAN, GAN and SMOTE augmentations were established on the 1: 50,000 stream sediment survey data, and used to automatically detect geochemical anomalies associated with polymetallic mineralization. The receiver operating characteristic (ROC) curve of the three ensemble models are very close to the upper left corner the of the ROC space, with the SMOTified-GAN augmented bagging ensemble model dominating the other two; and the area under the ROC curves (AUCs) of the three ensemble models are very close to 1.0 (0.99998, 0.99681, and 0.96803, respectively). Therefore, in terms of ROC curves and AUCs, the SMOTified-GAN augmented bagging ensemble model has the best performance in detecting geochemical anomalies associated with polymetallic mineralization. In addition, the geochemical anomalies associated with polymetallic mineralization detected by the SMOTified-GAN augmented bagging ensemble model have the close spatial correlation with the ore-forming control factors in the study area, and are mainly distributed around known polymetallic deposits. In other words, a bagging ensemble model with high performance can be constructed from the simple ELM classifiers with SMOTified-GAN augmentation in detecting geochemical anomalies associated with mineralization.