EXPLORATION OF MULTIVARIATE CHEMICAL-DATA BY PROJECTION PURSUIT

Automated laboratory procedures have made the acquisition and storage of large quantities of data possible. More measurements can be made on a single sample and more samples can be analyzed than was previously possible. However, it is often the case that these large quantities of data do not provide as much additional information as they might. This problem arises not because the additional data do not contain additional information, but because the data are in a form that does not make the information readily available. Substantial effort is required to identify the underlying structure of large data sets. Data having more than three variables become less comprehensible as the number of variables increases. Projection pursuit is a relatively new method for identifying lower-dimensional views of multivariate data by optimizing a criterion index that measures how interesting each projection is. Numerous indices have been proposed for use with projection pursuit. Several of these projection indices select projections based on the amount of clustering observed in each lower-dimensional view. Projection pursuit using one of these cluster-seeking indices has been applied to three sets of data to determine how the method performs when chemical and biological data are analyzed.