An efficient SVDD-base boundary method for multiclass classification problems

Abstract

In this article, we propose a new boundary method based on support vector domain description (SVDD) for multiclass classification problems. By utilizing the boundary around a class of data constructed by the SVDD, a new boundary distance measure, which gives the distance between a test object and the boundary around a class of data, can be determined by locating the nearest boundary point to the test object with the help of the geometrical interpretation of inner product. To decide which class the test object belongs to, a minimum distance classification is then carried out. Experimental results show that our method is robust to issues such as outliers and class imbalance, and overall it can achieve a comparable or better performance than other support vector-based classifiers for multiclass classification problems.

Keywords:

• multiclass classification
• support vector domain description
• minimum distance

Acknowledgments

This work was supported by the National Science Council of Taiwan (ROC), under Grant NSC98-2221-E-006-159-MY3.