A Hybrid GA and SA Algorithms for Feature Selection in Recogniti-libre

paper which submitted in
of 4
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
   y G          -p   Rea z , Bosh Pishgoo  Narges Noroi  Maryam koohadi  Fahimeh baesi   Dre Of Coe Alzha Uvesy Ten, Ia a@alzh Absrac-I th rsearch a hybrd feature electon technque baed on genec and mulated annng  algorthm  propoed. th approach  evaluated by  ung Bayean caier on a dataet of hand-pnted Far charactr whch nclude  ample for each  hand-prnted character. The acqured reult have been mproved by correcon of Smulated Annng through conderng two mnmum and mamum threhold. Kws: Getic Algh, Smued Aolng, pa recgi, feaure selec I. ITRODUCTIO I a ae ecogo syse, acevg a good eecveess s vey elaed o seleced feaes. ae ecogo dases re oe cracezed  y a lage e of eleva o edda feaes a ay sgcaly degade e ecogo esls nd edce e leg seed of algos, so adog a elevn sse of feaes s on o aage desoal colexy of s ole.  Deely, sg all exaced feaes, o oly as  o always e desale esl  also cease e e colexy of ecogo ocess. Geally, feae seleco s dg a sse of feaes wc ove e ecogo accacy. Ts cess as o a ases. Fs ase cldes a seac saegy o selec oe feae sse og all ossle, e secod ase cldes a eod fo evalag seleced sses w assgg a ess vae o e[]. Tee ae a lo of algos fo feae seleco.  geeal, we cn class feae seleco eods  ee gs. A s g, All ossle sses of colee feaes ae cosdeed ad evalaed fo eacg o a sale sse. A secod go, esc eods sc as fowrd nd ackad seleco ae adoed. I s go e algo ss s wok w a sse of feaes ad oe feaes ave ee eavely ad dmcally added 978-1-4244-6585-910$26.00 ©2010 EEE 4 o o oed o . Fally a d go, ado seac eods sc as geec algo nd slaed nealg eod c e sed . Refeece [2] as evalaed e eicecy of ese algos deals. Refeece [3] as dvded e feae seleco eod o dee coveg nd le eods o deede nd deede o of  vew o classes. Refeeces [4,5,6] ave evewed e eec of dffee feae seleco algos o e eecveess of ex classes nd cored e. I e ese rcle, ae feae exaco y cracesc loc eod, s we se geec algo nd e s cood spe y SA algo nd a e ed oved cood algo as o feae seleco eod. o class  s es s sle Bayesn. II. RE ROCESSIG Geeally,  age ocessg, e s se s cosdeed as e-ocessg. Ts cldes seveal sges wc ca e oed o o sln coeco,  oalzao ad g. Refeece [7] dels w evew of dee ecqes of oalzao nd e leeao. Refeece [8] oo desces e  ways owrds e leeao of ee eoed sges o scve nd-d of Frs . I s acle, Fas alaes as caaces re eg  oalzed olly ad eac crace s soded o oe fo sded ce. III. FEATURE EXTRACTIO I s acle e exsg feaes  ages ave  ee exaced og loc caacesc eod. Te ase of s eod s a e vecal nd ozol assg og eac age xel ca css e lack dos  fo ogal decos o 0, 1 o 2 os (  s ossle of cose a e es exceeds o,   s eod  deceases o o) . Teefoe  eac age xel we ca elae a fo   s e o a ee ass, so a eac  eeses e of vcal nd ozoal coss dos of a xel  oe of e a decos nd as a ao ewee 0 ad 2 . So we ca elae o evey eac xel a decal e eee 0 o 80. W s eod o eac ce 81 feaes ca  e exaced. e e elaed o feae I s eqal adaly of eeo of e of -1 eee decal es elaed o age xel. IV. FEATUR SELECTIO As evosly eoed  seco 1, s cn e claed a se of all exaced feaes o age,  o oly ceases e ae of calclao  also edces es ady ad always ca' gve e es ecogo ece. We as oof o, sly se all exaced feaes, nd a o classe y e. Te w es da evalao, we acqe e ece of ecogo  s case. e dog feae seleco og geec algo nd slaed aealg. Aewrds og e seleced of feae ses  s algos, classe ae aed aga nd evalae  y es daa. A e ed, esls ae eg coaed. A. Feature selectio through gnetic algorithm Gec algo s oe of e ado eods  wc ses gadal evolo eoy fo ole solvg . Oe of e a sse of s algo, keeg a se of es aswes  a olao. As   ology evolo eoy s algo as a ecns fo coosg es coosoes  a geeao. I s ocess cose coosoes  de oeao sc as Cossove nd Mao ae leeed. Ul ow  e eld of feae seleco og geec algo ee ave ee los of eserces doe.  efeece [9] fo ecogo of nd-ed crace ,oe aoac s eseed accodg o gec algo.  efeece [8] geec algo s sed as a eod of seleco sale feaes se, eded o ove sys ecogo of Frs crace nd-ed. I s acle, as eoed  evos seco, fo eac caace, 81 feaes re eg exaced. Te e of feaes as exaced re eg assed as  fo ease. Te gec algo sold selec o s  feae, sale sse fo classcao, so a e e of feaes ae  eg edced o esls of calclao nd also  ovde sale ecogo eceage o caaces classcao. Coosoes of s geec algo s a ay ge w a  leg s ake o cosdeaos. I oe wods eac coosoe as  gees. If e gee Beg oe, sows seleco ad f a eg zeo, sows o  seleco of feae aalogos w a gees. Fge 1 llsaes s coce.  I  I  FN    I   I  1 1 1   1 1         Fge  peseao of oe coosoe [6] Te al olao of s algo re seleced coleely ndo. I e calclaes ess fo eac coosoe w e sage of ess co. Fess co sed  s tcle s  fac e sme sle Bayesa classe a a eac e aed daa accodg o feaes of a coosoe ad calclaes ecogo eceage. I eac geeao, ee gos of coosoes ae eg asfeed o ex geeao. Te s go ae ose coosoes a ecogo ecege re fr oe n of a esold. Ts esold level s elaed o geao e nd as we go ge geeao, s ao ceases oo. Secod nd d gos re coosoes a esecvely de o coss ove ad ao o alf of coosoes of s go ad oe fo of oe ece geeao coosoes re ceaed accdeally. Te coss ove oeao, y coosg oe ndo ege e ewee 1 nd , nd e cage of coosoes als s doe o cossed do. Mao oeao oo s doe y dco of oe ndo e ewee 1 nd  ad cage of a gee o zeo  case of eg 1 ad vce vesa. Hee  s e oed a selecg feaes y  sg of geec algo s a vey e cosg oeao, ecase fo eac coosoe all daa s e aed nd evalaed a leas oce. B ally es smles wll e leeed as ad nd  w lesse eo. I s es, geec algo ae few geeao  wll aceve a ecogo eceage ge a  ry ecage. Te oale o s a e  e of feaes wc as ee evalaed y s eod ae edced accodgly. I e eod of loc caacesc, e e of feaes ae edced o 81 o 55. Resls re sow  ale 1. Geec algo as oe a ole nd a s,  alwas selecs coosoes w e es ecogo eceages, e ao nd coss ove oeao re efoed o e. Ts s we oe coosoe wc does o ave a sale ecogo eceage, e  ole obly g ave a vey g ecogo eceage we  xed w aoe cosoe. Oe sale dea fo adao of s ole s a coao  w slaed nealg algo wc wll e dscssed as follows. B Feature selectio through combiatio of Simulated Alig ad geetic algorithm Hee fo geec algo ole solvg, we coe  w SA algo. Te ocede s e sme. Hee oo sly we ceae a few accdeal coosoes ad calclae ecogo eceage og ess uco fo eac coosoe.  ee sead of ose coosoes wc e ecogo eceage re ge n esold ao wll e nsfeed o e ex geeao, all ece geeaos coosoes o e ass of e ecogo eceage, ave e cnce of  esece  e ex geao. I s cosdeed self evde a coosoe wc ave a lowe ecogo eceage, wll ave less cnce o e asfeed o ex geeao nd vce vesa.  e o case s a s cnce alog vey low,  s sll  exsece. coay o geec algo ose w low ecogo eceage ee s o a cace fo esece  ex geeao. Te oosed ocedes  coed algo SA ad GA, alog e sao s ceaed fo a sale coosoe seleco,  ee exss oe dececy. Sce ex geeao coosoes seleco s doe o obly ass, ee g e a codo a es coosoes of a olao  wold o ee o ex geao a all. Teefoe as  goes  s ossle a log e ae geeaos o ave ee coosoes,  s also ossle a e gadal feoes g e a. To esolve s ole  s eog o ave SA  e coeced a lle   ceao of ex geeao. Hee e oeao of coosoe seleco of ex geeao dog w o g ad low esolds. Fo ease,  ese o levels re aevaed o K nd B. If ecogo eceage fo a coosoe of low esold level K s lowe  ens a  wll eve ave e cnce of eg o ex geeao. vce  vesa If ecogo eceage fo a coosoe of g esold B s oe,  ens a  wll deely nsfeed o ex geeao. ow o all coosoes wc ave a ecogo ecage  eee ese wo K ad B esolds, ased o ecogo eceage e cnce wll e gve o e  ese  ex geeao. So e valale coosoes of ese geeao wll deely ave e ooy o e  ex geeao. Ad  vey valale coosoes of s geeao also  wll o e ese  ex geeao. B e es coosoes w a ecogo eceage wll e   oooae wll ave ossly o e es   ex geeao. By leo of s ecns we ca deve oe seul esls a e eoed o eods. V.  EVALUATIO ND COPARISO Te eseed ad aove eoed eods ave ee leeed nd evalaed y sg Bayesa classe o a daase of nded Frs caacs wc cldes 100 sales fo eac 33 ad-d caaces o Fas ad-ed craces . Oe fo of e oal exsg sales ave ee cosdeed as sale ess nd eag as ag sales. Te feaes of ese sales ave ee exaced og loc cracesc eod ae odced e ocesses. ,    } - )  G L 2) z· r )  ) / /  JI  9  d f  � ' J L   0.  ' J J  I Fue 2-sale se of -e Fas craces I e s s, oe sle Bayesn classe s aed w all 81 feaes ave ee exaced nd oseved a  es sage axaely 77 ece of sales ave ee classed coecly. I e secod se, we efoed feae seleco og accdel geec algo As od  seco 4.1.  s sae, geec algo y oed esde feaes ad seleced effecve nd sale feaes re efoed  addo o cease of ecogo ecage of craces l 80 ece, e e of feaes sed fo ag clsse oo s edced o 81 feaes o 55 .  I e d se,  ode o esolve e ole of geec algo, SA was added o e eod nd  fac oeao elaed o feae seleco og coed geec algo nd SA was  efoed . I s sa, algo w osso of esde feaes ad seleco of sale feae ses, ecogo eceage of caaces s ceased  ao 82 ece. I s case es of seleced feaes wee edced o 81 o 60 feaes. Tae I sows e esls. .... GA A+SA GAA with  resold 3 82 .... _ � - I ããã I � ã .   .          - 72  12 11 10 9 8 7 6 5 4 3   Fge 3. Coaave iges of oess  eceage classicao  ee geetos,  GA , GA+SA a  GA +SA ago w wo h a low esols Fge 3 gves a clea sg  elao w e  ogess of eoed algo  dee geeaos. As yo see, ose ges elaed o geec algo nd coed geec algo nd SA w o g ad low esolds ae ascedg. B ge elaed o coed geec algo Ad sle SA, re soees descedg nd soees ascedg . Ts vrao s e eec of oaly faco wc evosly was eoed  seco 4.2. I addo s corave ge exlas seoy of coed geec algo nd SA, w ow esolds  seleco sale feaes fo classcao. VI. SUMMARY Recogo syse of lees s ave g scy, g ady nd easy ools. Mny eods cold e exlaed a s e de  feae exaco nd classcao nd odce dffee esls. B exac all feaes s o always seul. I s acle fo edco of ole deso, wo geec ad coed geec nd SA algo  wee sed, nd  laly s ved a sage of all exaced feaes o oe age, fo classcao o oly colexy of calclaos ae ceased  also as always e ges ecogo   ecege s o ceaed. Teefoe edco of  ole deso sees ecessry og dee algo. Table 1-  Resls a Coaso Classifer Classifcato Feaure's rae umber N   77 8   y GA 80 55   y 82 60    GA d SA g w hgh d w hh EFERENCES [1] Ho-Duc Km, Cag-Hy  Pa, H-Cng Yg, Kee-Bo Geetc Algotm Bas Featue Selecto Metod Develomet fo Patte Recoto , peas  SICE-ICASE, Ieaoal Jo Cofeece   1020:1025,2006 [2] D.Zoge & A.Ja Algotms fo feate Selecto :n Evaluato , aeas : Pae Recogo, Poceegs of e 13 Iaoal Coece  volume2,   18.22,1996 [3] Jall saed, btaafa mi, cemet text clsscato efomace based move featue selecto metods ,volume40,  313:328, 2006. ( Fas) [4] Jaez Bak, Mako Gobelk,  Naaa Mlc-Faylg, Dja  Mldec Ieaco of Feae Seleco Meos n Lea Classcao Moels , Pceegs of e IC-02 Wokso o Tex Leg, 2002 [5] Aba Dasga, Peos Deas,Bolos Hab Feae Seleco Meos fo Tex Classcao , Ieaoal  Cofeece o Kowlege Dscovey a Daa Mg, Poceegs of e 13 ACM SIGKDD eaoal coece o Kowlege scovey a aa g,  230:239,2007 [6] Hqg L, Jya L, Lsoo Wog A Coaave Sy o Feae Seleco a Classcao Meos Usg  Gee Exesso Poles a Poeoc Paes , Geoe  Ifoacs 13,  51-60, 2002 [7] Cg-L Lu, Kaz Naasma, Hos Sko nd  Homc Fujsaa, Hnwtte dgt ecoto: vestgato of oalzato nd feat extcto tecques , Patte Recoto Socety . Publsed  by Elseve Scece B.V., Volume 37,  265279,  2004 [8] Kea amad Reza, hmnn esmael, otmzato of ecogto of Fas nwtg cracte based eectve feate selecto by  GA . 8  tellgce sy stem cofeece  Fedos uvesty ,  2007( Frs) [9] L.Coella, C.De Semo,F.Foaella a C.Mrocco A Feae Seleco Algo fo Hawe Caace  Recogo , aeas : Pae Recoo, ICPR 2008. 19 Ieaoal Cofeece,  1 :4,2008
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks