Online Hierarchical Streaming Feature Selection Based on Feature Interaction
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.
online streaming feature selectionhierarchical classificationfeature interactionsibling strategyneighborhood rough set
国家科技期刊平台
NSTL
万方数据
