Computer-Aided Cluster Analysis as a Rapid Method Assisting Mineral Identification

Long-Cheng Liang

Cluster analysis is one of the multivariate statistical methods that measure the similarity between objects. Aided by a microcomputer, cluster analysis offers rapid results that can lead to mineral identification based on the chemical composition of the minerals. In the process of cluster analysis, the chemical composition of the unknown minerals is either stored into a data file during an on-line quantitative analysis (e.g., electron microprobe analysis), or manually input to the data file. The data of the unknown minerals are respectively compared with an existing identification data matrix that consists of the chemical composition of related reference minerals. The unknown minerals can then be identified with the reference minerals by the closest similarity (e.g., high st correlation coefficient or lowest distance coefficient) in the dendrogram. By using the concentrations of major and minor elements as the variables, quantitative data obtained from some electron microprobe analyses have been tested and approximately 80% of unknown minerals are accurately identified.

Computer-aided cluster analysis not only accelerates the clustering process, but also provides a possible routine to identify minerals based on their chemical compositions. The accuracy of the identification can be improved if (1) error-prone raw data are discarded prior to the data inputting, (2) distortion effect generated in the clustering process is minimized, (3) selected reference minerals and unknown minerals belong to the same mineral series, and (4) x-ray diffractometry (XRD) is used to confirm the existence of the newly identified minerals in the sample.

AAPG Search and Discovery Article #91030©1988 AAPG Annual Convention, Houston, Texas, 20-23 March 1988.