Reduction from Cost-Sensitive Multiclass Classification to One-versus-One Binary Classification

Many real-world applications require varying costs for dierent types of mis-classication errors. Such a cost-sensitive classication setup can be very dierent from the regular classication one, especially in the multiclass case. Thus, traditional meta-algorithms for regular multiclass classication, such as the popular one-versus-one approach, may not always work well under the cost-sensitive classication setup. In this paper, we extend the one-versus-one approach to the eld of cost-sensitive classication. The extension is derived using a rigorous mathematical tool called the cost-transformation technique, and takes the original one-versus-one as a special case. Experimental results demonstrate that the proposed approach can achieve better performance in many cost-sensitive classication scenarios when compared with the original one-versus-one as well as existing cost-sensitive classication algorithms.