We have presented simple and general methods for obtaining accurate
class membership probability estimates for two-class and
multiclass problems, using binary classifiers that output ranking
scores. We have demonstrated experimentally that our methods
work well on a variety of data-mining domains and for different
classifier learning methods