Anna Bartkowiak
A Set of XLispStat Subroutines for Detecting Outliers
*****************************************************


A search for outliers using the grand tour method is performed. The grand tour
enables to inspect the configuration of the data points imagined as a data cloud
located in multivariate space. In the proposed method we rotate the data cloud
and inspect the projections in a scatterplot located in a 2D plane.

The algorithm works in two windows. In the first one (exhibiting  a plane)
the projections of the rotated points onto a 2D plane are displayed. An
ordinary or robust concentration ellipse with a given confidence level is
superimposed in the same plane on the displayed points. Point-projections which
fall outside the ellipse borders are suspected as outliers. In the second window
a linked count plot is steadily recording how often the individual data points
were notified outside the confidence borders.

In such way we obtain suspected outliers (points falling frequently
outside the borders of the confidence ellipse) and a clean data set constituted
from points which were always notified as located inside the confidence ellipse.

The procedure was tested using both benchmark data and some real data sets. The 
results of the tests were very satisfactory.