模糊核覆盖分类器及其应用  

Fuzzy kernel covering classifier and its application

作　　者：段震[1,2] 程家兴[1,2] 张铃[1,2] 

机构地区：[1]安徽大学计算智能与信号处理教育部重点实验室,合肥230039 [2]安徽大学计算机科学与技术学院,合肥230039

出　　处：《计算机应用研究》2010年第6期2043-2046,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60675031);安徽省高等学校优秀青年人才基金项目(2009SQRZ020ZD)

摘　　要：核覆盖算法是一种性能优秀的分类算法,但在拒识点处理方面存在不足。对核覆盖算法的构造过程进行了分析,修改了算法中覆盖半径的选取原则,对拒识样本引入隶属度函数,将算法推广为模糊核覆盖算法。讨论了孤立覆盖对分类器的影响,对覆盖数进行精简,降低计算量。通过实验验证改进算法的性能,并与其他模糊分类方法进行对比。将模糊核覆盖算法应用于垃圾邮件过滤,实验结果表明过滤器的性能得到了有效提高。While kernel covering algorithm （KCA） is a kind of effective classification algorithm, it still falls short in treating rejection points. This paper analyzed the construction process of kernel covering algorithm. By revising the selection criteria of covering radius and introducing the membership function for rejection points, generalized the algorithm as fuzzy kernel covering algorithm （FKCA）. Also discussed isolated covering’s impact on classifier’s performance and lowered the computation cost through reducing the number of coverings. Comparing with other classification methods on experiment results show this fuzzy kernel covering algorithm works well. FKCA is applied to spam filtering and the classifier’s performance is improved effectively.

关 键 词：覆盖算法 核函数 拒识样本 模糊核覆盖 垃圾邮件过滤 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]