KSVM错分及拒分区域的分类信息增益权重分类器

Information gain weight classifier for wrong and refused classification region of KSVM

为解决KSVM分类器错分及拒分区域问题,提出了一种新的结合分类信息增益权重的改进KSVM分类器(classifica-tion information gain weight KNN&&SVM,CIGWKSVM)。采用熵期望值度量训练样本的复杂程度、特征集针对分类的不确定性以计算特征集的分类信息增益值,并融合特征分布信息定义训练集样本各条件属性在分类过程中的CIGW权重。在此基础上,设计围绕加CIGW权的欧式距离测度进行聚类处理,并优化选择错分、拒分区K近邻代表点的CIGWKSVM分类器。从理论上比较分析了CIGWKSVM分类器的性能,仿真实验结果表明,CIGWKSVM分类器在保证效率的情况下,分类精度得到了极大的提高。

To deal with wrong and refused classification region for KSVM classifier,a new KSVM classifier that combined classification information gain weight is proposed.Entropy expectations is used to measure the complexity of training samples and the classification uncertainty of feature set for calculating classification information gain value,mix feature distribution to define CIGW of condition attributes on classification process.Then,CIGWKSVM classifier around Euclidean distance with CIGW weight is designed to cluster and select k neighbour representative point optimally.The performance of CIGWKSVM classifier is analyzed and compared in theory.The contrast simulation experiment shows that CIGWKSVM classifier raised the rate of classification accuracy greatly,meanwhile it also maintains efficiency of classification.