untitled

Size: px

Start display at page:

Download "untitled"

扎韶
4 years ago
Views:

ISSN 1000-9825, CODEN RUXUEW E-mal: jos@scas.ac.cn Journal of Software,2017,28(11):2891 2904 [do: 10.13328/j.cnk.jos.005350] http://www.jos.org.cn. Tel: +86-10-62562563,,, (, 100044) :, E-mal: zhhwang@bjtu.

1 ISSN , CODEN RUXUEW E-mal: Journal of Software,2017,28(11): [do: /j.cnk.jos ] Tel: ,,, (, ) :, E-mal: :.,,.,,., EPDS(Bayesan classfer algorthm based on emergng pattern for data stream).,.epds,.,. : ; ; ; : TP181 :,,,..,2017,28(11): : Du C, Wang ZH, Jang JJ, Sun YG. Bayesan classfer algorthm based on emergng pattern for data stream. Ruan Jan Xue Bao/Journal of Software, 2017,28(11): (n Chnese). Bayesan Classfer Algorthm Based on Emergng Pattern for Data Stream DU Chao, WANG Zh-Ha, JIANG Jng-Jng, SUN Yan-Ge (School of Computer Scence and Technology, Bejng Jaotong Unversty, Bejng , Chna) Abstract: Pattern-Based Bayesan model s one of the solutons for the classfcaton problem n data mnng. Most pattern-based Bayesan classfers consder the supports of patterns n the dataset of the home class only. However, the supports of the patterns n the counterpart class are gnored. In addton, for the hgh-speed dynamc changes and nfnte data stream, pattern-based Bayesan classfer whch ams at statc datasets can not work. To overcome these problems, EPDS (Bayesan classfer algorthm based on emergng pattern for data stream) s proposed. EPDS s a Bayesan classfcaton model based on the emergng patterns dscovered over data stream. In ths model, EPDS presents a smple hybrd forests (HYF) data structure to mantan the temsets of the transactons n memory, and uses a fast pattern extractng mechansm to accelerate the algorthm. EPDS adopts partally-lazy learnng strategy to update emergng temsets contnuously, and establshes a local classfcaton model n each class for the test transacton. Expermental results on real and synthetc data streams show that EPDS acheves hgher classfcaton accuracy compared to other classc classfers. Key words: data stream; emergng pattern; Bayesan; data mnng [1],., [2] :(1) (one-pass); : ( ) Foundaton tem: Natonal Natural Scence Foundaton of Chna ( ). : ; : , ; :

2 2892 Journal of Software Vol.28, No.11, Novermber 2017 (2) ;(3).,,,.,, X c ( ), X c ( ), X c c,., (emergng pattern, EP) [3] (Bayesan classfer algorthm based on emergng pattern for data stream, EPDS). [4,5]., [4]. EPDS. (partally-lazy learnng)., [6].EPDS,.,EPDS,., (fnd emergng pattern algorthm on data stream, FEP)., SFI-forest(summary frequent temset forest) [7] (hybrd-forest, HYF).. EPDS,,.,EPDS,. : (1),,, ; (2), ; (3),,. 1. 2,. 3 EPDS,. 4 EPDS ( ). DS=[T 1,T 2,,T n, ).,T n, DS n. 2( ). A,,.Ω A, A a (,a Ω) A,a. a A,a. A={A 1,A 2,,A n },Ω={Ω 1,Ω 2,...,Ω n } n. DS

3 : 2893 T td ={a 1,a 2,,a m },td,m T td,m n. T td 1. 3( ). DS X,X (Support(X)) DS X (Count(X)) DS ( DS ), (1). Count( X ) Support( X ) = (1) DS DS,.. 4( ). W n W +1 =[T n W +1,T n W +2,,T n ] DS W,, W,., W., X Count( X ) Support( X ) = (2) W,,,. 5( ). C DS,Ω C ={c 1,c 2,,c k },C W c, X (Support (X)) Count( X ) C= c Support ( ) X = (3) C, Count( X ) C= c C X. 6( ). DS W, X Support(X) (mnsupport), X,. FP={X Support(X) mnsupport}. 7( ) [3]. DS W, D 1 D 2, D 1 D 2 =,D 1 D 2 =D, X D 1 D 2 Growth(X,D 1,D 2 ) : 0, Support D ( X ) = Support 1 D ( X ) = 0 2, Support ( X ) 0, ( ) 0 D > SupportD X 2 1 = Growth( X, D1, D2) = Support D2( X ), SupportD 1( X ) 8( ). X, Growth(X,D,D 2 ) ρ, Growth(X,D 1,D 2 ) ρ, X D 1 D 2 ρ-ep, X D 2 EP.,D 2,D 1. c, C ; c, C. X C C Growth(X, C,C ). c Growth(X, C,C ) ρ,. 1.2,,. [8].,.,, NB(naïve Bayes),.NB P(T test,c ) (5). (4)

4 2894 Journal of Software Vol.28, No.11, Novermber 2017 n P( T, c ) = P( a, a,..., a, c ) P( c ) P( a c ) P( a c )... P( a c ) = P( c ) P( a c ) (5) test 1 2 n 1 2 n j j= 1,NB., NB., : [1,9], ;,, P(T test,c ) [10 13].,.. DS A 1,A 2,A 3,A 4,A 5 C.c,T test ={a 1,a 2,,a 5 }, (6) P(T test,c ) : P(T test,c )=P(a 1,a 2,,a 5,c ) P(c )P(a 1 c )P(a 2 a 1 c ) P(a 5 a 1 a 2 a 3 a 4 c ) (6). (7) P(T test,c ) : P(T test,c )=P(a 1,a 2,,a 5,c ) P(c )P(a 1 a 2 c )P(a 3 a 4 a 5 c ) (7), [9,14].,,.. [10,11].EnBay [10].EnBay. EnBay,. DSM-FI 7].,, SFI-forest,.,.MSW [15] SW-tree,. SW-tree,.., [5]., [16]. [17], [5,17,18].BCEP [19] MaxEPs [20].BCEP.MaxEPS..,.DFP-SEPSF [16],, [3],. 2 FEP,,,,. 2.1 FEP. SW=[T 1,T 2,,T SW ],, SW,T FEP, {1,2,..., SW }. T c C( C), HYF,,HYF={HYF }, c C., HYF. 3.,FEP,,

5 : 2895 HYF,,c C ;, (, ),FEP HYF HYF j (HYF j ), ;,,FEP, HYF={HYF },. HYF, DS c, HYF. HYF. (1) HYF={HYF }, {1,2,,k}. HYF 3 : (a lst of frequent tems, FI-lst) (frequent tem tree, FIT) (a lst of subsequence frequent tems, SFI-lst). (2) FI-lst c (, {1,2,,k}).FI-lst 3 :{x,x.count,x.head-lnk},,x,x.count x.x.head-lnk x.fit, FI-lst, x.fit x.sfi-lst. (3) x.fit x, 3 :{x,x.count,x.node-lnk},,x.count.x.node-lnk, x. (4) x.sfi-lst x, x x.fit.x.sfi-lst :{e,e.count},,e,e.count x.fit e. 1, HYF. 1 5, A 1,A 2,A 3,A 4,C,, C, 1 T d. SW =6. (1) HYF Table 1 Transacton dataset 1 T d A 1 A 2 A 3 A 4 C T 1 a 1 b 1 d 1 e 2 c 2 T 2 a 1 b 1 d 1 e 1 c 2 T 3 a 2 b 1 d 1 e 2 c 1 T 4 a 3 b 2 d 1 e 2 c 1 T 5 a 3 b 3 d 2 e 2 c 1 T 6 a 3 b 3 d 2 e 1 c 2 T 7 a 2 b 3 d 2 e 1 c 1 HYF. FEP T={a 1,a 2,,a n,c }, c C., T T, T ={a 1,a 2,,a n }., T,T a1 ={a 1,a 2,,a n },T a2 ={a 2,,a n },,T a(n 1) ={a n 1,a n },T an ={a n }, T a ={a,, a n }, {1,2,,n}.,FEP a FI-lst, T a a.fit, a.sfi_lst. 1,FEP 1 T 1 ={a 1,b 1,d 1,e 2,c 2 }, T 1 C 2., c 2 T 1 ={a 1,b 1,d 1,e 2 } 4 T a1 ={a 1,b 1,d 1,e 2 },T b1 ={b 1,d 1,e 2 },T d1 ={d 1,e 2 },T e2 ={e 2 }, 4 {a 1 },{b 1 },{d 1 },{e 2 } FI-lst, FIT, SFI_lst. 1 1 HYF 2. (2) HYF HYF :, DS,FEP

6 2896 Journal of Software Vol.28, No.11, Novermber 2017,. FEP 6,., c 1 C 1 {T 3,T 4,T 5 }, c 2 C 2 {T 1,T 2,T 6 }. 2 C 2 HYF 2. FI-lst FIT a 1 :1 b 1 :1 d 1 :1 e 2 :1 a 1 :1 b 1 :1 d 1 :1 e 2 :1 b 1 :1 d 1 :1 d 1 :1 e 2 :1 e 2 :1 e 2 :1 a 1.SFI-LIST a 1 :1 b 1 :1 d 1 :1 e 2 :1 b 1.SFI-LIST b 1 :1 d 1 :1 e 2 :1 d 1.SFI-LIST d 1 :1 e 2 :1 e 2.SFI-LIST e 2 :1 Fg.1 HYF 2 of C 2 after nsert T 1 1 T 1,C 2 HYF 2 FI-lst FIT a 1 :2 b 1 :2 d 1 :2 e 2 :1 a 1 :2 b 1 :2 d 1 :2 e 2 :1 b 1 :2 d 1 :2 d 1 :2 e 2 :1 e 2 :1 e 1 :2 e 1 :2 a 3 :1 a 3 :1 b 3 :1 d 2 :1 b 3 :1 d 2 :1 b 3 :1 d 2 :1 d 2 :1 a 1.SFI-LIST a 1 :2 b 1 :2 d 1 :2 e 2 :1 b 1.SFI-LIST b 1 :2 d 1 :2 e 2 :1 d 1.SFI-LIST d 1 :2 e 2 :1 e 2.SFI-LIST e 2 :1 e 1.SFI-LIST e 1 :2 a 3.SFI-LIST a 3 :1 b 3 :1 d 2 :1 b 3.SFI-LIST b 3 :1 d 2 :1 d 2.SFI-LIST d 2 :1 e 2 :1 Fg.2 HYF 2 of C 2 after nsert T 6 2 T 6,C 2 HYF 2,FEP,,. T old ={a 1,a 2,,a n,c }, HYF. HYF, n, FI-lst a 1,a 2,,a n 1, a.fit, {1,2,,n} 1, a.sfi_lst 1. 3 T 1 C 2 HYF 2. FI-lst a 1 :1 b 1 :1 d 1 :1 FIT a 1 :1 b 1 :1 d 1 :1 b 1 :1 d 1 :1 d 1 :1 e 1 :2 e 1 :2 a 3 :1 a 3 :1 b 3 :1 d 2 :1 b 3 :1 d 2 :1 a 1.SFI-LIST a 1 :1 b 1 :1 d 1 :1 b 1.SFI-LIST b 1 :1 d 1 :1 e 2 :1 b 3 :1 d 2 :1 d 1.SFI-LIST d 1 :1 e 1.SFI-LIST d 2 :1 a 3.SFI-LIST b 3.SFI-LIST d 2.SFI-LIST e 1 :2 a 3 :1 b 3 :1 d 2 :1 b 3 :1 d 2 :1 d 2 :1 Fg.3 HYF 2 of C 2 after sldng the wndow 3 C 2 HYF 2

7 : EPDS, :,, T test HYF ;, [3] ,EPDS T test HYF={HYF }.. DS, T test ={x 1,x 2,,x n };, HYF HYF.,. (1) T test ={x 1,x 2,,x n } x,, {1,2,...,n}, FI_lst ;, (2);, x +1 FI_lst, ; T test HYF FI_lst,. (2) FI_lst x x x.sfi_lst; x T test {x +1,x +2,, x n }, x.sfi_lst,, x {x,x j,,x k }, {x j,,x k } x.sfi_lst. (3) FI_lst x x x.fit; {x,x j,,x k } x.fit, {x,x j,,x k }, {x,x j,,x k }. {x,x j,,x k },,. (4) mn_sup {x,x j,,x k }, mn_sup,. (5),, (1)~ (4), , [3],. c, 3. (1) FP FP ; (2) FP FP ; (3) c EP. 3 (EPDS), ;,. 3.1 EPDS.,. (1)., DS A 1,A 2,A 3,A 4,A 5 C.c,T={a 1,a 2,,a 5 }, P(T,c ), HYF. {{a 1,a 2 },{a 3,a 4, a 5 }}, P(T,c ) P(c )P(a 1 a 2 c )P(a 3 a 4 a 5 c ). (2), [21]., C c 1,c 2. T,EPDS HYF 1 HYF 2 P(T,c 1 ),P(T,c 2 ).

8 2898 Journal of Software Vol.28, No.11, Novermber 2017 (3),., P(T,c ) P(T,c ) P(c )P(a 1 a 2 c )P(a 3 a 4 a 5 c ), P(a 1 a 2 c ) P(a 3 a 4 a 5 c ) {A 1,A 2 } {A 3,A 4,A 5 }, c,. : EPDS. (1).,.,. (2). (9), T C C, T c c. (3)..,,,. (4)., T c, P(T,c ),.,., (9). (5).,.,.,., ( ).,. Laplace [19]. 4(a) T={a 1,a 2,a 3,a 4,a 5,a 6 }. 4(b) T. 4(c) T., A 5 A 6., Laplace, P(a 5 a 6 c ) 1 Paa ( 5 6 c ) = count( c ) + attnum( A ) + attnum( A ) 5 6,attnum(A 5 ) attnum(a 6 ) A 5 A 6,count(c ) SW c. (8) C C C A 1 A 2 A 3 A 4 A 5 A 6 A 1 A 2 A 3 A 4 A 5 A 6 A 1 A 2 A 3 A 4 A 5 A 6 (a) (b) (c) Fg.4 Three cases of condtonally ndependent assumpton model , X. X c C c C ρ, X c C Support (X).,,,,,

9 : Support (X) P(X),,.,,. (1) X c C δ, X c C. (2) X Growth(X, C,C ) C Support (X), Suppor new (X) P(X). (9). Support ( X ) = Support ( X ) Growth( X, C, C ) (9) new,. 3.3 EPDS,,.,, FEP: Fnd Emergng Pattern Algorthm on Data Stream. : DS=[T 1,T 2,,T n, ), SW. : HYF={HYF },,. 01: for all T n DS 02: //w. 03: If w< SW 04: w=w+1; 05: HYFBuldng(T,HYF ); 06: end f 07: Else 08: w=w 1; 09: deletefromhyf(t old,hyf ); 10: w=w+1; 11: HYFBuldng(T,HYF ); 12: end f 13: end for, :,, ( 4 5 );,, ( 8 ~ 11 ). 3.4 EPDS,. T test, EPDS HYF,,. 2 EPDS. 2. ClassPredcton(HYF,T test,mn_sup,mn_sup j ). :HYF={HYF }, T test, c mn_sup, c mn_sup j. : T test c. 01: for all c n C

10 2900 Journal of Software Vol.28, No.11, Novermber : fnalitemset= ; 03: whle T test null do 04: // HYF c HYF j 05: FpatternC =selectfpitemset(t test,hyf,mn_sup ); 06: FpatternC j =selectfpitemset(t test,hyf j,mn_sup j ); 07: // 08: EPsets=MBD_LLBORDER(FpatternC j,fpatternc ); 09: // 10: bestfpattern=selectepset(epsets); 11: // 12: T test =T test /bestfpattern; 13: fnalitemset=fnalitemset bestfpattern; 14: end whle 15: //l fnalitemset 16: P(T test,c )=P(c )Π l fnalitemsetp(l c ); 17: end for 18: return the class c wth maxmal P(T test,c );,, ( 3 ~ 14 ). ( 16 )., ( 18 ). EPDS. m, O(m 2 ).EPDS.,.,EPDS,. 4. UCI. Massve Onlne Analyss(MOA) [22]. 3.60GHz Intel(R) Core(TM) CPU 8G Wndows UCI ( EEG,Frm MAGIC;5 Agrawal Generator [21] AGRAWAL Random RBF Generator RBF random RBF drft RBF Drft SEA Generator [23] SEA STAGGER Generator [14] STAGGER MOA Attrbute.,EEG,MAGIC,AGRAWAL,RBF,RBF Drft SEA. EPDS,. Weka( [24]. 2. Items -.,, 1,., 2,.AGRAWAL SEA

11 : Table 2 2 Real and synthetc data sets Dataset Transacton Attrbute Contnuous Nomnal Dscretzed Items Classes EEG Frm MAGIC AGRAWAL RBF RBF Drft SEA STAGGER (prequental error estmators) [25]., w w=10% n,,n ;,mn_sup =mn_sup j =0.01% w,,w.,,,. 4.3 EPDS, MOA. : Naïve Bayes(NB),Naïve Bayes Multnomal(NBM) [26] ; k-nn [27],k-NN wth PAW(PAW) [27] ; Rule Classfer(RC) [28],Rule Classfer Naïve Bayes(RCNB) [28] Hoeffdng Tree(HT) [29],Hoeffdng Opton Tree(HOT) [30],SGD,ORTO [31],FIMTDD [32]. 3., 2 ~ 12, 13 EPDS.. Table 3 3 Accuracy comparson wth other classfers Datasets NB NBM k-nn PAW RC RCNB HT HOT SGD ORTO FIMT-DD EPDS EEG Frm MAGIC AGRAWAL RBF RBF Drft SEA STAGGER Average ,EPDS. (1),EPDS STAGGER NB, NB NBM. (2),k-NN PAW EEG,RBF RBF Drft, EPDS k-nn PAW, k-nn PAW. (3),EPDS RC RCNB.

12 2902 Journal of Software Vol.28, No.11, Novermber 2017 (4),EPDS HT,HOT,SGD,ORTO FIMTDD. 3,ORTO FIMT-DD 3 RBF,RBF Drft SEA.ORTO FIMT-DD,. 3, ORTO FIMT-DD 3., MAGIC,NB. MAGIC, NB, EPDS NB,.,EPDS 3. (1),,,. (2),,,. (3),,. 3,,. 5 EPDS.EPDS, HYF.EPDS,,.,,.,EPDS.,.,,,. References: [1] Keogh EJ, Pazzan MJ. Learnng augmented Bayesan classfers: A comparson of dstrbuton-based and classfcaton-based approaches. In: Proc. of the 7th Int l Workshop on Artfcal Intellgence and Statstcs. Fort Lauderdale: Socety for Artfcal Intellgence and Statstcs, [2] Aggarwal CC. Data Streams: Models and Algorthms. Berln: Sprnger-Verlag, [3] Dong G, L J. Effcent mnng of emergng patterns: Dscoverng trends and dfferences. In: Proc. of the 5th Int l Conf. on Knowledge Dscovery and Data Mnng (KDD 99). San Dego: ACM Press, [do: / ] [4] Chen CC, Sh HX, Fan M. Aggregatng EP-based classfers to classfy data streams. Journal of Computer Research and Development, 2006,43(z3): (n Chnese wth Englsh abstract). [5] Sammut C, Webb GI. Encyclopeda of Machne Learnng and Data Mnng. Berln: Sprnger-Verlag, [do: / ] [6] Zhang P, Gao BJ, Zhu XQ, Guo L. Enablng fast lazy learnng for data streams. In: Proc. of 2011 IEEE the 11th Int l Conf. on Data Mnng (ICDM 2011). Vancouver: IEEE Computer Socety, [do: /ICDM ]

13 : 2903 [7] L HF, Shan MK, Lee SY. DSM-FI: An effcent algorthm for mnng frequent temsets n data streams. Knowledge and Informaton Systems, 2008,17(1): [do: /s ] [8] Lews DD. Nave (Bayes) at forty: The ndependence assumpton n nformaton retreval. In: Proc. of the 10th European Conf. on Machne Learnng (ECML-98). Berln, Hedelberg: Sprnger-Verlag, [do: /BFb ] [9] Fredman N, Goldszmdt M. Buldng classfers usng Bayesan networks. In: Proc. of the 13th Natonal Conf. on Artfcal Intellgence (IAAI 96). Portland: MIT Press, [10] Barals E, Caglero L, Garza P. Enbay: A novel pattern-based Bayesan classfer. IEEE Trans. on Knowledge and Data Engneerng, 2013,25(12): [do: /TKDE ] [11] Barals E, Caglero L. RIB: A robust temset-based Bayesan approach to classfcaton. Knowledge-Based Systems, 2014,71: [do: /j.knosys ] [12] Fan H, Ramamohanarao K. A Bayesan approach to use emergng patterns for classfcaton. In: Proc. of the 14th Australasan Database Conf. (ADC 2003). Adelade: Australan Computer Socety, [13] Meretaks D, Wjthrch B. Extendng nave Bayes classfers usng long tem sets. In: Proc. of the 5th ACM SIGKDD Int l Conf. on Knowledge Dscovery and Data Mnng (KDD 99). San Francsco: ACM Press, [do: / ] [14] Schlmmer JC, Granger RH. Incremental learnng from nosy data. Machne Learnng, 1986,1(3): [do: /A: ] [15] L GH, Chen H. Mnng the frequent patterns n an arbtrary sldng wndow over onlne data streams. Ruan Jan Xue Bao/Journal of Software, 2008,19(10): (n Chnese wth Englsh abstract). [do: /SP.J ] [16] Alav F, Hashem S. DFP-SEPSF: A dynamc frequent pattern tree to mne strong emergng patterns n streamwse features. Engneerng Applcatons of Artfcal Intellgence, 2015,37: [do: /j.engappa ] [17] García-Borroto M, Martínez-Trndad JF, Carrasco-Ochoa JA. A survey of emergng patterns for supervsed classfcaton. Artfcal Intellgence Revew, 2014,42(4): [do: /s x] [18] Lu Y, L JZ, Zhu JH. A novel graph classfcaton approach based on frequent closed emergng patterns. Journal of Computer Research and Development, 2007,44(7): (n Chnese wth Englsh abstract). [do: /crad ] [19] Domngos P, Pazzan M. On the optmalty of the smple Bayesan classfer under zero-one loss. Machne Learnng, 1997,29(2): [do: /A: ] [20] Wang Z, Fan H, Ramamohanarao K. Explotng maxmal emergng patterns for classfcaton. In: Proc. of the 17th Australan Jont Conf. on Artfcal Intellgence (AUS-AI 2004). Carns: Sprnger-Verlag, [do: / _102] [21] Agrawal R, Imelnsk T, Swam A. Database mnng: A performance perspectve. IEEE Trans. on Knowledge and Data Engneerng, 1993,5(6): [do: / ] [22] Bfet A, Holmes G, Krkby R, Pfahrnger B. MOA: Massve onlne analyss. Journal of Machne Learnng Research, 2010,11: [23] Street WN, Km YS. A streamng ensemble algorthm (SEA) for large-scale classfcaton. In: Proc. of the 7th ACM SIGKDD Int l Conf. on Knowledge Dscovery and Data Mnng (KDD 2001). San Francsco: ACM Press, [do: / ] [24] Fayyad UM, Iran KB. Mult-Interval dscretzaton of contnuous-valued attrbutes for classfcaton learnng. In: Proc. of the 13th Int l Jont Conf. on Artfcal Intellgence (IJCAI 93). Chambéry: Morgan Kaufmann Publshers, [25] Gama J, Sebastao R, Rodrgues PP. Issues n evaluaton of stream learnng algorthms. In: Proc. of the 15th ACM SIGKDD Int l Conf. on Knowledge Dscovery and Data Mnng (KDD 2009). Pars: ACM Press, [do: / ] [26] McCallum A, Ngam K. A comparson of event models for nave Bayes text classfcaton. In: Proc. of the AAAI-98 Workshop on Learnng for Text Categorzaton, Vol.752. Madson: AAAI Press, [27] Bfet A, Read J, Pfahrnger B, Holmes G. Effcent data stream classfcaton va probablstc adaptve wndows. In: Proc. of the 28th Annual ACM Symp. on Appled Computng (SAC 2013). Combra: ACM Press, [do: / ]

14 2904 Journal of Software Vol.28, No.11, Novermber 2017 [28] Gama J, Kosna P. Learnng decson rules from data streams. In: Proc. of the 22nd Int l Jont Conf. on Artfcal Intellgence (IJCAI 2011). Barcelona: AAAI Press, [29] Hulten G, Spencer L, Domngos P. Mnng tme-changng data streams. In: Proc. of the 7th ACM SIGKDD Int l Conf. on Knowledge Dscovery and Data Mnng (KDD 2001). San Francsco: ACM Press, [do: / ] [30] Pfahrnger B, Holmes G, Krkby R. New optons for hoeffdng trees. In: Proc. of the 20th Australasan Jont Conf. on Artfcal Intellgence (AUS-AI 2007). Gold Coast: Sprnger-Verlag, [do: / _11] [31] Zlobate I, Bfet A, Pfahrnger B, Holmes G. Actve learnng wth evolvng streamng data. In: Proc. of the European Conf. on Machne Learnng and Prncples and Practce of Knowledge Dscovery n Databases (ECML PKDD 2011). Athens: Sprnger- Verlag, [do: / _39] [32] Ikonomovska E, Gama J, Džerosk S. Learnng model trees from evolvng data streams. Data Mnng and Knowledge Dscovery, 2011,23(1): [do: /s y] : [4],,. EP.,2006,43(z3): [15],..,2008,19(10): htm [do: /SP.J ] [18],,..,2007,44(7): [do: / crad ] (1983 ),,,, (1992 ),,,,,,.. (1963 ),,,,,CCF,,. (1982 ),,,,CCF,,.

中文模板

中文模板 ISSN 000-9825, CODEN RUXUEW E-mal jos@scasaccn Journal of Software, Vol8, No4, Aprl 2007, pp905 98 http//wwwjosorgcn DOI 0360/jos80905 Tel/Fax +86-0-62562563 2007 by Journal of Software All rghts reserved