Exploring outliers
Here we use a dataset from the Palomar-QUEST sky survey to look for outliers. The dataset consists of 1000 matched objects in a small region. The data are from
two successive nights using Gunn rizz and Johnson UBRI filters. We concentrate on the various colors.
- Boxplot is very good at revealing the relationships between the different colors, including the mean, median, the overlap and the outliers for the set
- Hierarchichal clustering reveals the clustering present in the data.
- A follow-up run with desired number of clusters to cut the dendogram
provides a list of outliers along with cluster centers and withinss.
- Kmeans allows a visual inspection of the different clusters in terms
of all the desired parameter pairs.
- K-Density associates a probability with each object about its being
an outlier.We sorted the probabilities dividing the set into 4 zones and
plotted them in these color-color plots. The scatter is clear even though
we see here only 2 out of 9 dimensions
- Two sets of B R I images:
the object at the center of the top set is from the main locus.
The object in the second set is one of the outliers - clearly due
to its relatively unusual I brightness.
