Use Of Simulation Data Sets For Assessing Interlaboratory Pattern Recognition Accuracy

Abstract

Interlaboratory comparisons using common (reference) materials of known composition are an established means for assessing overall measurement precision and accuracy. Intercomparisons based on common data sets are equally important and revealing when one is dealing with complex chemical patterns or spectra requiring significant numerical modeling and manipulation for component identification and quantification. Two case studies of chemo-metric using simulation test data (STD) are presented, one comprising STD vectors as applied to nuclear spectrometry and the other STD data matrices as applied to aerosol source apportionment. Generic information gained from these two exercises included (a) the requisites for a successful STD intercomparison (including the nature and preparation of the simulation test patterns); (b) surprising degrees of bias and imprecision associated with the data evaluation process per se; (c) the need for increased attention to implicit assumptions and adequate statements of uncertainty; and (d) the importance of STD beyond the intercomparison, i.e., their value as a research tool for improv-ing the state of the pattern recognition art.