Use of Ensembles for Outlier Overcoming

A method for overcoming the influence of outlier samples in projection methods is described in this chapter, having been devised and tested as a proof of concept for the use of ensembles in combination with unsupervised learning.

The main goal of all visualization techniques is to bring to the user a deeper understanding of a multi-dimensional data set by generating some kind of graphical representation that can be easily inspected by the naked eye, enabling the viewer to rapidly focus on the most interesting groups or clusters of data. Projection methods are those based on the identification of “interesting” directions in terms of any specific index and/or projection. These indexes or projections are, for example, based on the identification of directions that account for the largest variance of a data set, by using methods such as Principal Component Analysis (PCA) (see Section 2.4.1) (Pearson, 1901; Hotelling, 1933), or by looking for higher order statistics, such as the skweness or kurtosis -which is the case of Exploratory Projection Pursuit (EPP) (Friedman, 1987)- .