摘要
为降低特征维数并提高分类准确率,提出一种基于ReliefF算法、互信息和类可分性法的多评价准则融合特征选择方法。利用序关系分析法确定3种评价准则的重要性权值系数,按照多评价准则融合模型获得特征重要性排序,通过支持向量机分类器实现最终特征选择。通过3个UCI标准数据集进行仿真实验,实验结果表明,和单准则的特征选择方法相比,该方法在保证良好鲁棒性的基础上,能够有效降低特征维数,具有更高的分类准确率。
To reduce the feature dimension and improve the classification accuracy,a feature selection method based on ReliefF algorithm,mutual information and class separability method was proposed.The importance weights of the three evaluation criteria were determined using the sequence relation analysis method,and the feature importance ranking was obtained according to the multi-evaluation criterion fusion model,and the final feature selection was realized using the support vector machine classifier.Simulation experiments were carried out on three UCI standard datasets.The simulation results show that compared with the single criterion selection method,the proposed method can effectively reduce the feature dimension and have higher classification accuracy on the basis of guaranteeing good robustness.
作者
于宁宁
刘刚
刘森
曹冰许
YU Ning-ning,LIU Gang,LIU Sen,CAO Bing-xu(College of Information Engineering,Henan University of Science and Technology,Luoyang 471023,Chin)
出处
《计算机工程与设计》
北大核心
2018年第7期2075-2079,共5页
Computer Engineering and Design
基金
河南省自然科学基金项目(162300410095)
河南省重点科技攻关基金项目(172102210039)
关键词
特征选择方法
多评价准则融合
RELIEFF算法
互信息
类可分性法
序关系分析
feature selection
multiple evaluation criteria
ReliefF algorithm
mutual information
interclass divisibility me- thod
sequence relation analysis method
