Experimental results show that ensembles can produce better results than any
classifier in isolation. Bell et al. [11], for instance, used a combination of 107 differ-
ent methods in their progress prize winning solution to the Netflix challenge. They
state that their findings show that it pays off more to find substantially different ap-
proaches rather than focusing on refining a particular technique. In order to blend
the results from the ensembles they use a linear regression approach and to derive
weights for each classifier, they partition the test dataset into 15 different bins and
derive unique coefficients for each of the bins. Different uses of ensembles in the
context of the Netflix prize can be tracked in other approaches such as in Schclar et
al.’s [67] or Toescher et al.’s [71].