基于模糊邻域判别指数的在线流组特征选择

Online group streaming feature selection based on fuzzy neighborhood discrimination index

在线流组特征选择可以充分利用特征流中原始的组结构信息,以在线的方式处理特征选择问题。然而,现有方法大多无法处理具有模糊性和不确定性的数据。为此,提出一种基于模糊邻域判别指数的在线流组特征选择算法。设计一种模糊邻域判别指数,用于描述模糊邻域粒的判别信息,扩展相关的不确定性度量方法。在此基础上,用组内特征选择和组间特征选择两种策略选择具有强近似能力且非冗余的特征。在8个公共数据集上进行对比实验,验证了该算法具有更优且稳定的分类性能。

Online group streaming feature selection can make full use of the original group structure information in the feature stream to handle the feature selection problem in an online manner.However,most of the existing methods cannot handle the data with ambiguity and uncertainty.To this end,an online group streaming feature selection algorithm based on fuzzy neighborhood discrimination index was proposed.A fuzzy neighborhood discrimination index was designed to describe the discriminant information of fuzzy neighborhood granules and extend the related uncertainty measures.On this basis,two strategies,intra-group feature selection and inter-group feature selection,were used to select features with strong approximation ability and non-redundancy.The comparative experiments on eight public datasets verify that the algorithm has better and stable classification performance.