Verification and interpretation of statistical clusters of geochemical elements using self-organizing maps

In geochemical data analysis, chemical elements are often clustered by statistical means first to reveal any established association of some elements due to similar originality, colocation, chemical bonding, or contamination. Statistical clustering ranks a predominately single association among elements very highly and the double or multiple associations of elements lowly according to their correlation coefficients. These lowly ranked double/multiple associations of elements in fact may also be useful in identifying new relationships among the associated elements. We propose a new intuitive method to fast verify statistical clusters of geochemical elements using self-organizing maps (SOM) of neural networks. SOM clustering offers an independent approach that can not only reaffirm the highly ranked statistical clusters of elements, but also further classify those lowly ranked statistical clusters into double or multiple associations of elements.