Detección de anomalías geoquímicas para plata (Ag) y arsénico (As) utilizando agrupamiento jerárquico (hierarchical clustering) y bosques aislados (isolation forest) en datos de geoquímica de superficie

Abstract

Abstract: Modern geological exploration requires advanced methodologies for the detection of mineral deposits, overcoming the limitations of univariate statistics. This study presents a hybrid Machine Learning approach for the vectorization of Silver and Arsenic anomalies. A database of 4,000 surface geochemistry samples was analyzed. To mitigate the closure effect and spurious correlations of compositional data, the original matrix was subjected to a clr transformation. Subsequently, Hierarchical Clustering was applied, which allowed the delineation of three background clusters. Singularity detection was executed using the Isolation Forest algorithm with a 5% contamination rate, generating a continuous Anomaly Score. Geospatial evaluation in GIS environments demonstrated that Silver Kriging omits important targets, while Arsenic Kriging generates false positives on the western flank. In contrast, the Anomaly Score achieved absolute agreement with the discrete anomalous vectors in the northeastern sector, demonstrating the superiority of the multivariate algorithm in filtering and delineating exploration targets with high precision.