周华平 张贝贝 刘光宗

摘 要：倾斜数据集在实际应用中具有很高的价值，同时也是数据挖掘中的难点和热点，因而得到越来越多的研究者的关注。通过给出的一种改进的支持向量机—DP-SVM，首先对数据集进行预处理、分类，再对它们采取相应的操作；然后根据混叠数据集中样本的k个最近邻的类别，并结合波动大小与其相应类限定值的关系，对该样本点做相应的处理；最后根据副类支持向量与主类样本数量之间的关系，决定对副类支持向量采取何种修剪策略，训练得到其分类器。提出的分界修剪支持向量机由消除分类边界混叠与支持向量修剪两个主要处理部件构成，并通过这两个处理部件实现对分类边界混叠数据的判别及两类支持向量的修剪。实验结果表明，DP-SVM是一种能够较好地解决倾斜数据集分类边界混叠问题的有效方法。

关键词：倾斜数据集；混叠现象；支持向量；分界修剪支持向量机

中图分类号： TP391 文献标志码：A

文章编号：1672-1098（2017）06-0008-07

Abstract：This paper was to present an improved Support Vector Machine-DP-SVM. Firstly data sets were preprocessed and classified， and then the corresponding operation were taken on them. Secondly， according to the category of the k nearest neighbors in aliasing data set and the relationship between the fluctuation and the corresponding category limited values， the sample point was processed correspondingly. Lastly， based on the relationship between the support vectors in negative category and the number of samples in major category， the pruning strategy was determined on the support vectors in negative category，as a result， the classifier was trained to be obtained. The proposed DP-SVM in this paper was composed of two main processing components， with one eliminating aliasing data at the classifying boundary and the other supporting the vectors pruning； and the two components could be used to realize the judgement of the aliasing data at the classifying boundary and support the vectors prunning of the two categories. The results show that DP-SVM is an effective method that can better solve the aliasing problem of skewed data sets.

Key words：skewed data sets； aliasing phenomenon； support vector； demarcation pruning-support vector machine

傾斜数据集在实际应用中具有很高的价值，同时也是数据挖掘中的难点和热点，因而得到越来越多的研究者的关注。对倾斜数据集的处理，文献[1-2]提出了许多处理效果较好的理论方法，但其主要是对分类边界混叠数据做了简单的删除或对支持向量做了简单修剪，而未考虑这些数据对数据集分类正确率及主类分类精度的影响；因此对存在倾斜的数据集的分类正确率及主类的分类精度的处理效率尚有进一步提升的空间。……