GEOFCM: a new method for statistical classification of geochemical data using spatial contextual information

Abstract

Conventional clustering algorithms such as k–means and fuzzy c–means (FCM) cluster analysis do not fully utilize the spatial distribution information of geologic samples. In this paper, we propose GEOFCM, a new clustering method for geochemical datasets with location coordinates. A spatial FCM algorithm originally constructed for image segmentation was modified for application to a sparse and unequally–spaced dataset. The proposed algorithm evaluates the membership function of each sample using neighboring samples as a weighting function. To test the proposed algorithm, a synthetic dataset was analyzed by several hyper–parameter settings. Applying this algorithm to a geochemical dataset of granitoids in the Ina–Mikawa district of the Ryoke belt shows that samples collected from the same geological unit are likely to be classified as the same cluster. Moreover, overlapping geochemical trends are classified consistently with spatial distribution, and the result is more robust against noise addition than standard FCM analysis. The proposed method is a powerful tool to use with geological datasets with location coordinates, which are becoming increasingly available, and can help find overviews of complicated multidimensional data structure.