Research On Online Hierarchical Streaming Feature Selection Algorithm Based On Decision Error Rate

As a momentous data pre-processing method of machine learning and artificial intelligence,feature selection can select some significant features from the original data to reduce dimensions of data,and improve the model learning capability.However,in real-world life,feature space is usually accompanied by unknown and dynamic,and features exist in the form of a stream.Traditional feature selection methods assume that all features are available at once,without considering the trend of features changing over time.In addition,data also contains rich hierarchical relationships.Traditional feature selection methods fail to consider the structural relationships between data.These emerge a problem that traditional feature selection algorithm is unable to meet many requirements when dealing with high-dimensional data and unknown feature space.Motivated by these,based on streaming feature environment,the research focus on feature selection for data with a hierarchical structure.The main research works as follows:(1)Online hierarchical streaming feature selection based on neighborhood decision error rate.Based on the achievements and problems of existing traditional online streaming feature selection method,and considering the hierarchical relationships between data,a new online streaming feature selection model for hierarchical structured data is presented.The model fully considers the sibling relationship of the hierarchical of classes,proposes two online evaluation criteria,including online significance selection and online relevance analysis,and designs the corresponding feature selection algorithm.Experimental studies show that the proposed algorithm can achieve better performance than traditional feature selection methods based on hierarchical data.(2)System for online streaming feature selection.On the basis of the algorithm research,a system of online streaming feature selection is designed and implemented by using the MATLAB.First,the system implements two modules,including feature selection module and evaluating indicator module.The system includes traditional streaming feature selection algorithms,feature selection algorithms for hierarchical structure data and evaluating indicator of algorithms.Then,the function of the system is explained and tested through experiments.The system is installable,visual,and easy to operate,and also has practical application value.