Abstract:

Support vector data description (SVDD) is a data classification algorithm of one-class data description, which has the minimum structure risk and attracts much attention recently. The SVDD performance of classification results is determined by the parameter optimization. As the sample point density is defined, a parameter optimization function based on minimizing the density outside is proposed. The proposed algorithm can avoid the calculation of miss detection rate during the optimization, and make full use of sample data distribution information to improve the SVDD performance. Compared with the UCI database, the simulation results confirm that the parameter optimization algorithm can reduce the miss detection rate and miss classification rate effectively.