A “Weighted” Geochemical Variable Classification Method Based on Latent Variables

Clustering of variables relies on relationships among them. The strength of those relationships is generally measured by the correlation coefficients between pairs of variables. This paper proposes specified variable weighted correlation coefficients and takes the clustering around latent variables (CLV) approach as an example to transform the common clustering method into a “weighted” clustering method. The aim is to eliminate factors that are unrelated to the variable that was adopted for weighting to ensure that the cluster centers are sufficiently different and have good correlations with the adopted variable. A log-transformed dataset was used to evaluate the proposed method. Three clusters were obtained under the restriction of the As element, and they represented three ore-controlling factors related to the Goldenville Formation, namely geologic features such as formation, fault contacts, and granitoid intrusions. Not only did the new cluster centers account for most of the variability related to the weighted element (As) but they also showed significant differences in spatial distributions.