Information entropy based dynamic data stream classification model
Guo Yan-fen · Jisuanji gongcheng yu sheji · 2014
Due to concept drift would make the classifier unstable,to improve classification accuracy and anti-concept-drifting ability,an information entropy and classifier pool mechanism based data stream classification model is proposed.The proposed model employs sliding window method converting the dynamic data stream into static data block form firstly.Then,the proposed model employs the information entropy algorithm to detect concept drift problem.If the concept drift detected,the current classifier would be updated,otherwise the current classifier would not be changed.In addition,for recurrent concept drift situation, apool mechanism,which reserves the historical concept and related classifier,is employed for classifier selection.The experiment employs multiple datasets for validation,and multiple classifiers are used for comparing experiment.