一种基于粗糙集属性约简的多分类器集成方法

为提高多分类器系统的分类精度,提出了一种基于粗糙集属性约简的分类器集成方法 MCS_ARS。该方法利用粗糙集属性约简和数据子集划分方法获得若干个特征约简子集和数据子集,并据此训练基分类器;然后利用分类结果相似性得到验证集的若干个预测类别;最后利用多数投票法得到验证集的最终类别。利用UCI标准数据集对方法 MCS_ARS的性能进行测试。实验结果表明,相较于经典的集成方法,方法 MCS_ARS可以获得更高的分类准确率和稳定性。

New classifier ensemble method based on rough set attribute reduction

To improve the accuracy of multiple classifier system, this paper proposed an classifier ensemble method MCS_ARS. This method used rough set attribute reduction and data partition to obtain a number of features subset and data subset to train base classifier, then it used the similarity of the classification results to get the results of validation set and got the final classification results of validation set by majority voting. Experiment results on UCI data sets show that compared to classical ensemble methods, MCS_ARS has higher classification accuracy and stability.