ide this question empirically on a case-by-case basis.
The predominant approach to choose classifiers empirically is to estimate
the accuracy of candidate algorithms on the problem, usually via crossvalidation1
, and select the one which seems to be most accurate. Schaffer (1993) has
investigated this approach in a small study with three learning algorithms on
five UCI datasets. His conclusions are that on the one hand this procedure is
on average better than working with a single learning algorithm, but, on the
other hand, the crossvalidation procedure often picks the wrong base algorithm
on individual problems. This problem is expected to become more severe with
an increasing number of classifiers.