Rotation-based outlier detection for geochemical anomaly identification in stream sediment multivariate data

Abstract This research explores the use of the rotation-based outlier detection (ROD) method for identifying geochemical anomalies in a multivariate stream sediment dataset from Iran, targeting porphyry and vein-type Cu mineralization. Geochemical datasets often present challenges for outlier detection methods like local outlier factor (LOF) and k-nearest neighbor (KNN), which rely on distance or density metrics and require parameter tuning (e.g., neighborhood size k). High-dimensional feature spaces further complicate their application. ROD, in contrast, offers a parameter-free, rotation-based approach that effectively analyzes geometric relationships between samples in subspaces, mitigating the curse of dimensionality. This makes ROD particularly suited to high-dimensional geochemical datasets, where complex relationships between elements (due to lithology or mineralization) are critical for identifying anomalies. This study compares ROD with LOF and KNN using two subsets of geochemical variables (Ag, As, Au, Bi, Co, Cr, Cu, Mo, Ni, Pb, Sb, Zn; and Ag, As, Au, Cu, Mo, Sb) and evaluates its performance based on the receiver operating characteristic (ROC) analysis and the number of known mineral occurrences detected in anomaly class. ROD outperforms LOF and KNN, capturing 78% (14 out of 18) of known Cu-bearing mineral occurrences. Moreover, ROD shows better conformity between 10% of highest outlier scores and Cu-mineralization sites. Rotation cost function in ROD, evaluated using the median absolute deviation (MAD), enhances its ability to detect outliers by focusing on orientation rather than distance, and by reducing noise misclassification. In addition, the parameter-free design of ROD and improved handling of high-dimensional data makes it a promising tool for geochemical exploration, as it captures unique mineralization-related signals that might be missed by traditional methods.