3:30 PM - 3:45 PM
[SGC54-07] A new statistical method to identify geochemical data structure
Keywords:multivariate statistical analysis, cluster analysis, principal component analysis, independent component analysis, geochemical data
Here we propose a new multivariate statistical method that combines three conventional but powerful methods to capture the true structure of multivariate data [Iwamori et al., 2017, doi:10.1002/2016gc006663]; they are k-means cluster analysis (KCA), principal component analysis (PCA), and independent component analysis (ICA). The reasons for selecting the three methods are (i) KCA and PCA are probably the most fundamental yet powerful tools for multivariate analyses; (ii) ICA is not as common as PCA but is a unique tool for identifying hidden independent structures; and (iii) the three methods are newly found to be closely related and can be integrated to analyze the data effectively. In this study, we first describe the relationship of these three methods to elucidate the entire data structure based mainly on synthetic data. We apply this to a natural data set of isotopic compositions of basalts for which ICA has been performed. On the basis of the results, an effective combination of the methods is clarified, for which we provide an Excel program “KCA” at both doi:10.1002/2016gc006663 and http://dsap.jamstec.go.jp/ to allow readers to test and apply the program to individual problems.