COMPARISON AND EVALUATION OF HIERARCHICAL CLUSTER TECHNIQUES APPLIED TO AUTOMATED ELECTRON-PROBE X-RAY-MICROANALYSIS DATA

For the processing of the large amount of data generated by automated electron probe x-ray microanalysis (EPXMA) of particulate samples, hierarchical cluster analysis are invoked. To evaluate the performance of seven hierarchical cluster techniques, cluster analyses were performed on a number of known mineral mixtures and the degree of correct classification was measured quantitatively by kappa statistics. In addition to the evaluation of the different cluster techniques, the influence on the cluster result of a number of experimental parameters was determined. In practice, ten mineral combinations were studied as a function of the mineral mixture ratio, using both normalized and unnormalized variables. For this purpose a total of 7000 cluster and kappa analyses were performed. In general, it seemed that Ward's method was most successful in finding the correct classification. For one mineral mixture, 2100 results were studied to elucidate the effect of the mixture size, of working with correlated/ uncorrelated variables (principal component space) and of the quantification of the EPXMA data (different deconvolution techniques applied to quantitative data obtained by the Armstrong-Buseck ZAF corrections).