不平衡支持向量机的调整方法  

Adjusting Method for Imbalanced Support Vector Machines

作　　者：刘万里[1,2] 刘三阳[2] 王金艳[3] 

机构地区：[1]洛阳师范学院数学系,洛阳471022 [2]西安电子科技大学应用数学系,西安710071 [3]洛阳师范学院物理系,洛阳471022

出　　处：《计算机科学》2009年第3期148-149,155,共3页Computer Science

基　　金：国家自然科学基金项目(60574075)资助

摘　　要：针对两类不平衡数据的分离超平面的偏移问题提出一种调整方法。首先对两类样本数据进行初步训练,产生一个初步分离超平面的法向量,然后把高维样本投影到该法向量上得到一维数据,然后由一维投影数据所提供的信息利用非参数密度估计,根据误差率最小准则,得到一个修正的偏置。几个真实数据试验表明,该方法不仅能保持标准支持向量机的良好性能,而且能平衡或减少总的错分率。An adjusting method for the offset of separating hyperplane of binary-classification imbalanced data was pro- posed. Firstly, the two class of samples were trained preliminarily and a normal vector of the separating hyperplane was generated. Secondly one dimensional data were obtained by projecting the high dimensional sample data onto the normal vector. Finally a new algorithm was proposed by using the information provided by the estimation of the non-parameter density for the projection data. Thus the offset of the separating hyperplane of the standard support vector machines （SVM） was revised. Several real experiments show the proposed method can keep the good performance of standard SVM and balance or decrease the whole error ratio.

关 键 词：不平衡数据 投影 支持向量机 Parzen窗密度 偏置 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]