One-class support vector machine with compositional data analysis to recognize geochemical anomaly patterns related to mineralization

The accurate delineation of mineralization-related geochemical anomalies is recognized as a fundamental challenge in greenfield exploration, particularly where labeled data are limited and geological complexity is pronounced. In this study, the One-Class Support Vector Machine (OCSVM) is integrated with Compositional Data Analysis (CoDA) to enhance the recognition of subtle multivariate anomalies associated with gold mineralization. To address the closure problem inherent in geochemical datasets, isometric log-ratio (ILR) and centered log-ratio (CLR) transformations are applied to project compositional data into Euclidean space prior to modeling. The OCSVM algorithm, implemented with a radial basis function (RBF) kernel, is trained on original, CLR, and ILR-transformed datasets to evaluate the influence of compositional preprocessing on anomaly detection performance. Anomaly scores generated by each model are spatially mapped and validated using linear productivity (LP) indices derived from borehole drilling data. Model performance is quantitatively assessed using receiver operating characteristic (ROC) analysis and the corresponding area under the curve (AUC). Results from 604 lithogeochemical samples and 57 drillholes referred to the study of anomalies in presence of gold mineralization in the Khunik prospect region, southeastern Iran indicate that ILR-transformed data achieve the highest predictive performance (AUC = 0.708), followed by CLR (AUC = 0.704) and raw data (AUC = 0.635), emphasizing the importance of appropriate compositional treatment in multivariate anomaly modeling. Moreover, controlled noise is injected into the different geochemical datasets to assess the robustness of anomaly detection results in this approach. The integration of CoDA and OCSVM is shown to provide a robust, data-driven framework for geochemical anomaly detection, offering enhanced sensitivity and specificity for mineral exploration targeting in complex geological settings.