資源描述:
《一種基于重取樣的代價(jià)敏感學(xué)習(xí)算法_谷瓊》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫。
1、CN43-1258/TP計(jì)算機(jī)工程與科學(xué)2011年第33卷第9期ISSN1007-130XCOMPUTERENGINEERING&SCIENCEVol.33,No.9,2011文章編號(hào):1007-130X(2011)09-0130-06*一種基于重取樣的代價(jià)敏感學(xué)習(xí)算法ANovelCostSensitiveLearningAlgorithmBasedonRe-sampling谷瓊,袁磊,寧彬,熊啟軍,華麗,李文新GUQiong,YUANLei,NINGBin,XIONGQi-jun,HUALi
2、,LIWen-xin(襄樊學(xué)院數(shù)學(xué)與計(jì)算機(jī)科學(xué)學(xué)院,湖北襄陽441053)(SchoolofMathematicsandComputerScience,XiangfanUniversity,Xiangyang441053,China)摘要:大多數(shù)非均衡數(shù)據(jù)集的研究集中于純重構(gòu)數(shù)據(jù)集或者純代價(jià)敏感學(xué)習(xí),本文針對(duì)數(shù)據(jù)集類分布非均衡和不相等誤分類代價(jià)往往同時(shí)發(fā)生這一事實(shí),提出了一種以最小誤分類代價(jià)為目標(biāo)的基于混合重取樣的代價(jià)敏感學(xué)習(xí)算法。該算法將兩種不同類型解決方案有機(jī)地融合在一起,先用樣本類空間重構(gòu)的方法使原始數(shù)據(jù)集的兩類數(shù)據(jù)達(dá)到基本均衡,然后再引入代價(jià)敏感學(xué)習(xí)算法進(jìn)行分類,能提高少數(shù)類分類精
3、度,同時(shí)有效降低總的誤分類代價(jià)。實(shí)驗(yàn)結(jié)果驗(yàn)證了該算法在處理非均衡類問題時(shí)比傳統(tǒng)算法要優(yōu)越。Abstract:Moststudiesontheimbalanceddatasetclassificationfocusonthediscussionofre-sam-plingorcost-sensitivelearningsystemsthemselves;however,thefactthatthecostsofimbalancedclassdistributionandunequalmisclassificationerrorsalwaysoccursimultaneouslyisnegle
4、cted.Weproposeanovelcostsensitivelearning(CSL)algorithmwhichcombinesthemethodsofre-samplingandtheCSLtechniquestogetherinordertosolvethemisclassificationproblemofimbalanceddataset.Ononehand,there-samplingtechniqueallowsthebalanceddatasetsbyreconstructingboththemajorityandthemi-norityclass.Ontheothe
5、rhand,theclassificationisperformedbasedontheminimalmisclassificationcostbutnotthemaximalaccuracy.Herethemisclassificationcostfortheminorityclassismuchhigherthanthemisclassificationcostforthemajorityclass.Acost-sensitivelearningprocedureisthenconduc-tedforclassification.Theexperimentalresultsshowth
6、attheproposedmethodcanimprovetheclassifi-cationaccuracyanddecreasethemisclassificationcosteffectively,andthealgorithmissuperiortothetraditionalalgorithmsasfordealingwiththeimbalancedproblem.關(guān)鍵詞:分類;非均衡數(shù)據(jù)集;混合重取樣;代價(jià)敏感學(xué)習(xí)Keywords:classification;imbalanceddataset;hybridre-sampling;costsensitivelearningd
7、oi:10.3969/j.issn.1007-130X.2011.09.023中圖分類號(hào):TP301.6文獻(xiàn)標(biāo)識(shí)碼:A*收稿日期:2011-05-20;修訂日期:2011-07-26基金項(xiàng)目:國(guó)家自然科學(xué)基金資助項(xiàng)目(61075063);國(guó)家863計(jì)劃資助項(xiàng)目(2009AA12Z117);湖北省自然科學(xué)基金資助項(xiàng)目(2010CDB05201);湖北省教育廳中青年項(xiàng)目(Q20112604)通訊地址:441053湖北省襄陽市