Mapping geochemical anomalies using angle-based outlier detection approach
Year: 2024
Type: article
Abstract: This study assesses the effectiveness of the angle-based outlier detection (ABOD) method in identifying geochemical anomalies within the complex geological setting of the Ahar-Arasbaran Zone (AAZ) in the Alborz-Azerbaijan Magmatic Belt (AAMB). This region is known for its diverse mineralization, including Cu-Mo porphyry deposits, epithermal base and precious metal veins, and Fe-Cu skarn deposits. A dataset comprising 1067 stream sediment samples, collected in 2007 and subjected to centered log-ratio transformation to manage compositional data limitations, served as the basis for this investigation. In this study, the impact of dimensionality on the ABOD method was investigated by implementing both the original ABOD and its approximation, FastABOD, which considers a certain number of neighboring samples. The study revealed that selecting key elemental commodities and pathfinders led to more effective anomaly detection compared to utilizing all 32 trace and minor elements. As the number of input variables decreased, the differences between ABOD and FastABOD scores diminished, emphasizing the influence of dimensionality on anomaly detection performance. Furthermore, the efficacy of four ABOD implementations (i.e., ABOD_All, ABOD_Selective, FastABOD_All, FastABOD_Selective) in delineating geochemical anomalies was assessed. Results indicated that ABOD_Selective captured around 61% of known mineral occurrences in the highest score class (Q4), ranking highest in this aspect. FastABOD_Selective performed well in integrating Q3 and Q4 classes, indicating its effectiveness in anomaly detection.Receiver operating characteristic (ROC) curves were utilized in the research to assess the effectiveness of the ABOD method, comparing its performance using the locations of known mineral occurrences and anomaly-checking stage rock samples. Notably, FastABOD_All and ABOD_Selective demonstrated slightly superior performance when known deposits were used as labeled data. When mineralized samples were included in the analysis, the selective approach exhibited superior performance, further confirming its efficacy in prediction-area analysis. Additionally, elemental contributions to anomaly scores produced by different ABOD variations were investigated using random forest regression. Key elements such as Sb, Au, As, and Cu were found to play significant roles in delineating geochemical signals from various mineralization types and processes. Moreover, a comparative analysis with independent component analysis (ICA) was conducted, focusing on independent components like IC2 and IC5, representing their effectiveness in outlining porphyry Cu-Mo deposits and skarn mineralization, respectively. The study demonstrated the ability of ICA to highlight unique geochemical patterns in mineralized zones. In outlining mineralization sources, especially those enriched in Cu and Au, IC2 was superior in highlighting Cu-enriched zones, while the ABOD method effectively delineated both Au- and Cu-bearing zones, providing valuable insights for future mineral exploration endeavors in complex geological settings like the AAZ.
Authors: Shahed Shahrestani, Ioan Sanislav
Institution: James Cook University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,730
OpenAlex ID: W4404530314