However, in many applications, a ranking of examples according
to class membership probability is not enough. What is needed is
an accurate estimate of the probability that each test example is a
member of the class of interest.
Probability estimates are important when the classification outputs
are not used in isolation but are combined with othersources of
information for decision-making, such as example-dependent misclassification
costs [25] or the outputs of another classifier. For
example, in handwritten character recognition the outputs from the
classifier are used as input to a high-level system which incorporates
domain information, such as a language model