Abstract: The rough set theory is an important mathematical tool to deal with uncertainty and incompleteness. This paper proposed a novel rough oncclass support vector machine by introducing rough margin into oncclass support vector machine. With the definitions of upper approximation and lower approximation hyperplanes, the influences of training samples on the decision hyperplane arc determined adaptively by their position within the rough margin. Moreover, outlier samples are prone to produce small margin errors since they lie close to the upper approximation hyperplane, so that the overfilling problem of decision hyperplane can be avoided. Experimental results on UCI datasets show the superior generalization performance of rough oncclass support vector machine.

Key words: Rough set, One-class, Support vector machine