Mapping geochemical anomalies using angle-based outlier detection approach

This study evaluates the effectiveness of the angle-based outlier detection (ABOD) method in identifying geochemical anomalies in the Ahar-Arasbaran Zone (AAZ) within the Alborz-Azerbaijan Magmatic Belt (AAMB), known for its diverse mineralization such as Cu-Mo porphyry deposits, epithermal base and precious metal veins, and Fe-Cu skarn deposits. ABOD and its fast approximation (FastABOD) were applied to datasets with 9 (Selective) and 32 (All) features. Results showed that with fewer variables, the performance difference between ABOD and FastABOD decreased, highlighting the impact of dimensionality on anomaly detection. Geochemical anomaly maps (ABOD_All, ABOD_Selective, FastABOD_All, FastABOD_Selective) were assessed for detecting known mineralization. ABOD_Selective demonstrated superior performance, effectively placing 76 % of skarn and 70 % of porphyry mineral occurrences into the highest anomaly class, despite its overall performance being approximately 61 %. Additionally, ABOD was compared with independent component analysis (ICA), focusing on IC2 and IC5. ICA effectively highlighted unique geochemical patterns, with IC2 excelling in identifying Cu-enriched zones and ABOD effectively delineating both Au- and Cu-bearing zones. ABOD was also compared with local outlier detection methods LOF, kNN, and iNNE. LOF showed distinct anomaly distributions due to its local density approach, while kNN, iNNE, and FastABOD produced similar maps based on distance, isolation, and angle-distance. ROC analysis revealed no significant performance difference, though FastABOD showed slight superiority, particularly with mineralized samples as validation points. Moreover, applying principal component analysis (PCA) as feature selection method enhanced the performance of the FastABOD method in delineating geochemical anomalies related to hydrothermal mineralization. Finally, random forest regression identified key elements such as Sb, Au, As, and Cu as significant in distinguishing geochemical signals from various mineralization types. • ABOD with key elements effectively detects anomalies in high-dimensional data. • ABOD_Selective captures a significant proportion of known mineral occurrences. • ICA, especially IC2 and IC5, delineates porphyry Cu-Mo and skarn mineralization. • ABOD detects high-dimensional outliers well but faces high computational costs.