Explaining a presence of groups in analytical data in terms of original variables

Abstract

This manuscript shows the usefulness of Projection Pursuit (PP) and Multivariate Regression Trees (MRT) for analytical data exploration. Additionally, features of Projection Pursuit and kurtosis as a projection index are presented. The ability of Projection Pursuit to discover groups in the data is compared to classical Principal Component Analysis (PCA). Moreover, it is also demonstrated how the presence of groups in the data can be explained in terms of explanatory variables with the aid of Projection Pursuit and Multivariate Regression Trees. Neither Projection Pursuit nor Multivariate Regression Trees are commonly used for exploring chemical data however, they are able to enrich to a high extent the interpretation.