基于特征交互的层次分类在线流特征选择

Online Hierarchical Streaming Feature SelectionBased on Feature Interaction

在开放动态环境下的分类学习任务中,数据特征空间具有动态性,标记空间存在层次化结构.现有的层次分类在线流特征选择算法可以选择较优的特征子集,但这些算法忽略了特征之间存在的交互作用.基于此,提出了一种基于特征交互的层次分类在线流特征选择算法.首先,设计了一种基于层次邻域依赖度去判断特征交互的计算方法;其次,针对层次化结构数据,根据层次结构中不同节点间的兄弟关系定义邻域粗糙集模型;最后,设计了具有在线重要性分析、在线冗余性分析以及在线交互性分析的层次分类在线流框架,用于选择强相关和存在交互作用的特征子集.在6个层次数据集上的实验验证了所提算法具有较优的综合性能.

In classification learning tasks in open dynamic environments,the data feature space is dynamic and there is a hierarchical structure in the labelling space.Existing hierarchical classification online streaming feature selection algorithms can select a superior subset of features,but these algorithms ignore the interactions that exist between the features.Therefore,this paper proposes a feature selection algorithm for hierarchical classification online streaming based on feature interaction.Firstly,a computational method based on hierarchical neighborhood dependency is designed to judge the feature interaction.Secondly,for hierarchical structure data,a neighborhood rough set model is defined on the basis of sibling relationships between different nodes in the hierarchical structure.Finally,the online streaming framework is designed for hierarchical classification with online importance analysis,online redundancy analysis and online interaction analysis for selecting the subset of features that are strongly correlated and have interaction.The proposed algorithm is experimentally verified on six hierarchical datasets to have superior comprehensive performance.