摘要
由于缺少对数据结构信息的考虑,现有的域描述型单类分类器得到的支撑面往往是次优解。因此,以支持向量数据描述(SVDD)算法为基础,通过一种简易的形式引入数据亲和因子以保持样本局部特性,提出保局性数据域描述分类器(LPDD),使成簇的数据作用被强化,而呈零星分布的数据影响力被削弱,引导分类支撑面自动靠近数据高密区而提高算法性能。此外,为适应大样本应用场合,采用序列最小优化算法进行模型参数调整。实验证明,所提算法无论在训练速率还是在分类性能上都优于SVDD。
In a support vector data description（SVDD）,the compact description of target data was given in a hyper spherical model which was determined by a small portion of data called support vectors.Despite the usefulness of the conventional SVDD,however,it may not identify the optimal solution of target description due to neglecting the structure of the given data.In order to mitigate this problem,a novel one-class-classifier named locality preserving data domain description（LPDD） was proposed which takes the data density into account by using of affine factor.Besides,the sequential minimal optimization was adopted to adjust model parameters for applying in the large sample occasions.Experiments with various real data sets show promising results.
出处
《计算机科学》
CSCD
北大核心
2011年第11期208-212,共5页
Computer Science
基金
国家自然科学基金(61070043)资助
关键词
亲和因子
支持向量域描述
序列最小优化
单类分类器
Affine factor
Support vector domain description
Sequential minimal optimization
One-class-classifier
