A performance study of local outlier detection methods for mineral exploration with geochemical compositional data

Abstract

In exploration geochemistry, mineral deposits are typically characterised by an enrichment of the targeted elements, and thus their element composition differs from that of samples in a local neighbourhood. Local outlier detection methods aim at identifying local changes. In contrast to conventional outlier detection procedures, local outlier detection methods are multivariate methods for outlier identification that incorporate the spatial neighbourhood of the samples. It is essential that geochemical data are treated as compositional data, and the requirements for their treatment depend on the specific local outlier detection method. We demonstrate how prominent local outlier detection methods can be used for mineral exploration with geochemical data that vary in scale, in the sampling density, and in data quality. The methods are compared based on known mineralisations, and recommendations for their usefulness are provided.