基于代表的交叉验证分类

Representative-based cross validation classification

基于代表的邻域覆盖粗糙集分类算法,在某些数据集上表现良好,数据的类别不平衡问题严重影响算法的分类精度。为尽量消除类别不平衡问题的影响,在k折交叉验证方法的基础上,针对基于代表的邻域覆盖粗糙集分类算法,提出了3种集成策略。策略1依靠k折交叉验证,获得对应的k个基分类器,所有的基分类器组成委员会对未分类样本分类;在策略1的基础上,策略2选择分类精度相对较高的基分类器组成委员会,对未分类的样本进行分类;策略3在前2种策略的基础上,利用主动学习的思想,对训练集进行扩充,得到新的分类器再对未分类样本分类。实验所用数据集为UCI标准数据集,且对k的取值做了对比实验。结果显示,3种策略均有不同程度的提升,且k取5时总能取得较好的提升效果。对于不同数据集,应选择相适应的改进策略。

Representative-based classification through covering-based neighborhood rough sets,it performs well on some data sets,and the imbalance of data categories seriously affects the classification accuracy of the algorithm.In order to eliminate the impact of category imbalance as much as possible,based on the k-fold cross-validation method,this paper proposes three ensemble strategies for the representative-based classification through covering-based neighborhood rough sets algorithm.The first strategy relies on k-fold cross-validation to obtain corresponding k base classifiers,and all base classifiers form a committee to classify unclassified samples.On the basis of the first strategy,the second strategy selects a base classifier with relatively high classification accuracy to form a committee,and then classifies the unclassified samples.Based on the first two strategies,the third strategy uses the idea of active learning to expand the training set to obtain a new classifier and then classify the unclassified samples.The data set used in the experiment is the UCI standard data set,and a comparative experiment has been done on the value of k.The results show that the three strategies have different degrees of improvement,and when k is set to 5,a better improvement effect can always be achieved.For different data sets,appropriate improvement strategies should be selected.