JISE

[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19]

Journal of Information Science and Engineering, Vol. 30 No. 6, pp. 1985-2002

Support Vector Domain Description with Maximum Between Spheres Separability

MOHAMED EL BOUJNOUNI¹, MOHAMED JEDRA² AND NOUREDDINE ZAHID³
Faculty of Sciences
Laboratory of Conception and Systems (Microelectronic and Informatics)
Mohammed v – Agdal University
B.P. 1014, Rabat, Morocco
E-mail: med_elbouj@yahoo.fr¹; jedra@fsr.ac.ma²; zahid@fsr.ac.ma³

Support Vector Domain Description (SVDD) is inspired by the Support Vector Classifier. It obtains a sphere shaped decision boundary with minimal volume around a dataset. This data description can be used for novelty or outlier detection. Our approach is always to minimize the volume of the sphere describing the dataset, while at the same time maximize the separability between the spheres. To build such sphere we only need to solve a convex optimization problem that can be efficiently solved with the existing software packages, simulation results on seventeen benchmark datasets have successfully validated the effectiveness of the proposed method.

Keywords: support vector domain description, small hypersphere, maximum separability

Retrieve PDF document (JISE_201406_17.pdf)