Stat 640 Alternate Final Dataset

Astronomers are cataloguing all of the objects in the
sky these days.  One group (ESO) uses the telescope
on La Silla, Chile, to image the so-called
Chandra Deep Field South.  An example image is

   http://www.mpia.de/COMBO/combo_index.html

The dataset consists of 3,438 galaxies in the CDFS
for which all measurements are available.  We
are interested in the speed at which these galaxies
are receding from ours.  The 5th variable is commonly
used as a surrogate for this purpose, being the mean
red-shift.  Variables 2-7 are related, so should not
be used as predictors.  (Variable 1 is just a pointer
to a larger database.)

Now a histogram of variable 5 shows a strong bimodal
distribution.  So we define Y = I( x5 > 0.45 ) as
an indicator of which cluster a galaxy falls.  Can
we find a good prediction algorithm from variables
8-30?

Variables BjMAG to S280MAG measure absolute magnitudes
of the galaxy in 10 bands.  (The telescope uses 17
different filters to measure different parts of the
spectrum.)  The remaining variables measure the
observed brightness in 13 bands in sequence from 420 nm
to 915 nm (ultraviolet to far red).

One could also predict the magnitude of Y directly,
as an alternative goal.