基于局部类相似的特征选择方法

Local Class Similarity Based Feature Selection Method

常用的特征选择方法利用样本空间的整个区域提取最优的特征子集.与此相反,提出一种新的局部特征选择方法,即样本空间的每个区域都与各自不同的最优特征集相关联,这些特征集能够最优地适应样本空间的局部变化.同时,在求解最优特征集对应的子空间时,基于最近邻思想,提出了一种度量测试数据与各个类相似性的方法,用来对测试样本进行分类.提出的方法可以描述为线性规划优化问题,因此可以通过简单的凸优化来求解全局最优解.在3组真实数据集和3个主流的方法上进行的对比实验,结果证明了该算法的可行性和有效性.

The common feature selection methods utilize the entire region of the sample space tOextract an optimal subset of features.In contrast,this paper proposes a new local feature selection method,in which each region of the sample space is associated with a different optimal feature set,which can optimally adapt tOthe local variation of the sample space.Based on the concept of the nearest neighbor research,this paper proposes a method ineroduced tOmeasure the similarity between the test datas sOas tOclassify the tested samples.The method ineroduced in this paper can be described as the problem in a linear programming optimization,sOthe global optimal solution can be realized by simple convex optimization.The experimental results are gained by comparing the three real datasets and three mainstream methods demonstrate the feasibility and effectiveness of the proposed algorithm.