资源描述：
《关于模糊决策树生成过程中启发式算法研究》由会员上传分享，免费在线阅读，更多相关内容在工程资料-天天文库。
1、AbstractFuzzydecisiontreelearningisaveryefficientandpracticalmethodonaccountofmergingtheadvantagesoffuzzysetstheoryanddecisiontreelearning.Ithasbeensuccessfullyappliedtoabroadrangeofknowledgeacquisitionsfromdiagnosemedicaltoassesstocreditriskofloanapplic
2、ant.However,constructingtheoptimalfuzzydecisiontreehasbeenprovedtobeNP-hard,aheuristicalgorithmisnecessary.FUZZY—ID3algorithmisthemostpopularandthemostefficientheuristicalgorithmintheexistingreferences,theheuristicinformationinwhichisfuzzyinformationgain
3、,i.e.,thefuzzymutualinformationbetweentheattributesandtheclass.Itmakesthefuzzyinformationentropyconvergequickly,andwecanobtainasmalltreewithoutmuchcomputation.Butalmostallheuristicalgorithmsignoretherelevance(redundancy)betweenattributes.Therefore,aftera
4、nalyzingtheeffectofrelevancebetweenattributesonthegenerationoffuzzydecisiontreeandtheshortcomingofFUZZY—ID3indetail,weproposeanextendedalgorithmtoFUZZY—ID3.Anupdatedheuristicisadapted,whichaimsatselectingtheattributethatcanbringnotonlythemoremutualinfonn
5、ationbetweenacandidateattributeandtheclass,butalsothelessmutualinformationbetweenacandidateattributeandtheselectedattributesonthesamebranchasthenexttestingattribute.Theextensionisbasedonaddingtheprocessofreducingtheeffectofredundancybetweenattributestoge
6、nerationoffuzzydecisiontree.NumerousfiguresandtablesfromtheexperimentdataarecitedtoillustratethedifferencebetweentheextendedheuristicalgorithmandFUZZY-ID3.Theresultturnsouttobetwosituations:oneisthat,tosomedatasets,theextendedalgorithmissuperiortoFuzzy・I
7、D3,andinanothersituation,thetwoareequalKeyword:MachineLearning;DecisionTreeLearning;FuzzyDecisionTree;HeuristicAlgorithm;MutualInformation第一章序言§1」背景知识学习是人类具冇的一种重要智能行为,那么3机器能否像人类一样也能具有学习能力呢？这就是人工智能学科中的一个重要领域一机器学习所研究的内容刈机器学习稍为严格的定义［1］是机隣学习是一门研究机器获取的学新知识和新技
8、能，并识别现有知识的学科Q其实，早在人工智能成为一个独立的学科Z前:就已经开始了对机器学习的研究o从最初的基于神经元模型以及函数逼近论的方法研究,到以符号演算为基础的规则学习和决策树学习的产生，和Z后的认知心理学小归纳、解释、类比等概念的引入，至最新的计算学习理论和统计学习的兴起当然还包括基于马尔可夫过程的增强学习5机器学习一直都在和关学科的实践应用中起着主导作用［2］不仅如此机器学习还是人工智能中最前沿的研究课题Z—,它一直被公认为是设计
