Twin-SVDD Classifier with the Conception of Relative Distance

摘要：

Twin support vector domain description(Twin-SVDD) classifier with the conception of relative distance was proposed in this paper. The preliminary SVDD model described the target dataset with one class by constructing a compact optimized hypersphere in feature space. And the model was effective to deal with problems of pattern classification with imbalanced dataset such as outlier detection. But only information about the positive class of dataset was used in the preliminary SVDD model. As for binary classification problem, inspired by the construction of Twin support vector machine where nonparallel planes were solved separately, the Twin-SVDD model was proposed. Two optimized hyperspheres which described positive and negative class of datasets were constructed separately in the Twin-SVDD model. So information about both classes of dataset was used. And then new classification decision-making function was constructed based on the parameters of the Twin-SVDD model with the conception of relative distance. At last, experiments were performed.Experimental results showed that the TwinSVDD model was more effective than the preliminary SVDD model when dealing with pattern classification problems.And the proved classification ecision-making function improved the performance of the Twin-SVDD model.

关键词： support vector domain description twin support vector domain description relative distance pattern classification

作者: Lei Yang Wei-Min Ma Bo Tian

作者单位: School of Economics and Management Tongji University 1239 Si-Ping Road, Shanghai, China School of Information Management & Engineering Shanghai University of Finance Economic 777 Guo-ding

会议类型: 国际会议

会议名称: The 10th International Conference on Intelligent Technologies(第十届智慧科技国际会议 InTech09)

会议地点: 桂林

会议语种:英文

页码: 554-561

在线出版日期: 2009-12-12（万方平台首次上网日期，不代表论文的发表时间）