Cost-sensitive learning and decision making revisited

In many real-life decision making situations the default assumption of equal misclassification costs underlying pattern recognition techniques is most likely violated. Then, cost-sensitive learning and decision making bring help for making cost–benefit-wise optimal decisions. This paper brings an up-to-date overview of several methods that aim to make a broad variety of error-based learners cost-sensitive. More specifically, we revisit direct minimum expected cost classification, MetaCost, over- and undersampling, and cost-sensitive boosting.