一种能够适应概念漂移变化数据流分类方法.pdfVIP

2 4 V ol . 2 . 4 2007 8 CA A I T r an sactio ns on Int el lig ent Sy st em s A ug . 2007 富春岩, 葛茂松 ( 佳木斯大学 公共计算机教研部, 黑龙江 佳木斯 154007) , , . , , , . ; ; ; ; T P311. 13 A 2007) A data stream classification methods adaptive to concept drift F U Chunyan, GE M aoson g ( Commo nality T eaching Dep artm ent of Computer, Jiamusi U niv ersity , J iamu si 154007 , China) Abstract: A t present , most clas sificat ion m et hods f or data st ream s are dev elo ped w it h t he as sumpt ion of st eady dat a dist ribut ion . H ow ever, t he dat a coll ected fro m t he r eal w orl d w ill ch ange over a p eriod of t ime in t he under lying co ncept s ( know n as concept drift ing) . T his low ers the pr edict ive precisio n of a clas sifica t ion m odel. T his p aper pro poses a classificat ion algo rit hm t h at can ident if y and adapt to occurrences of concept dr if t ing according t o t he ch aract er ist ics o f t he dat a st ream . Exp eriment s show t hat t he proposed algorit hm dynamically adju st s t he size of t he t raining w indow and t he number of new ex amples during model recon str uction acco rding t o t he current rate of co ncept drift ing . Keywords: dat a st ream s; clas sificat ion ; concept drif t ing; o nline lear ning; decisio n t ree . , , ( dr if t ing concept ) , , . , . , ID 3[ 1] C4 . 5[ 2] , CA RT [ 3] . , ,