PROJECTION PURSUIT VERSUS RELATIVE PROJECTION PURSUIT

摘要

This paper deals with an extension of projection pursuit named relative projection pursuit comparing with conventional projection pursuit. Projection pursuit is a powerful computer intensive method for finding "interesting structures" in high dimensional data. In the conventional projection pursuit, the definition of "interesting" is that is far from the normal distribution, because the normal distribution the most natural distribution and has been studied for a long time. But in the case that the purpose on the analysis is to investigate a feature of a subset of whole data set, we would like to search the projected direction that the projected distribution of the subset is far from the distribution of the whole data set. We have developed relative projection pursuit, whose purpose is to find out projections that reveal non-normal structures in the data set. We will show a numerical example to compare projection pursuit and relative projection pursuit.