Maximal-Margin Approach for Cost-Sensitive Learning Based on Scaled Convex Hull

摘要

In this paper, a new maximal margin method, scaled convex hull (SCH) method is proposed to solve the cost-sensitive learning. By providing different SCH with a different scale factor, the initial overlapping SCHs can be reduced to become separable, and the existing methods can be used to find the separating hyperplane. The new method changes the distribution of the sample, which assigns different scale factor. The experiment results are used to validate the effectiveness of the scaled convex hull and its simplicity.