Research on One-class Classifier Algorithm Based on Closed-loop Tree Constrained by Minimum Mean Value Covering Model

Traditional One-class Classifier based on covering model is difficult to describe the main structure and specific edge of data manifold accurately at the same time. So this paper proposes a new space covering model-closed-loop tree tructure. It uses minimum balanced tree model to fit the main manifold of data, and uses closed chain model to link or enfold the multi-branch structure of local edge, forming a smooth, clear the edge of covering model. The Two models match each other by the constraint principle of minimizing mean, and integrated as a whole. It can describe the main structure and specific edge of data manifold effectively at the same time. The results of experiments based on several databases prove that the proposed method is good.