《数据挖掘、机器学习和Weka》VIP免费

数据挖掘—实用机器学习技术及Java实现原书–英文版《DataMining—PracticalMachineLearningToolsandTechniqueswithJavaImplementations》，新西兰IanH.Witten、EibeFrank著Weka–AnopensourceframeworkfortextanalysisimplementedinJavathatisbeingdevelopedattheUniversityofWaikatoinNewZealand.–http://www.cs.waikato.ac.nz/ml/weka/–http://www.mkp.com/datamining/概念：KDD、ML、OLAP与DMKDD（KnowledgeDiscoveryinDatabase）是一种知识发现的一连串过程。ML（MachineLearning）＝KD，不限于Database的数据过程：挖掘－数据模式－表示－验证－预测OLAP（OnlineAnalyticalProcess）是数据库在线分析过程。数据挖掘（dataMining）只是KDD/ML的一个重要组成部分。DM用在产生假设，而OLAP则用于查证假设概念：DM与DBDataPreparation要占Datamining过程70％工作量「Database」＋「Datamining」＝会说话的数据库概念：DataMining概念：数据挖掘是从大量的数据中，抽取出潜在的、有价值的知识（模型或规则）的过程–KeyCharacteristicsofDataMining：•Largeamountofdata•Discoveringpreviouslyunknown,hiddeninformation•Extractingvaluableinformation•MakingimportantbusinessdecisionusingtheinformationDM/ML的一些要点–Thedataisstoredelectronicallyandthesearchisautomatedbycomputer;–Aboutsolvingproblemsbyanalyzingdataalreadypresentindatabases;–Definedastheprocessofdiscoveringpatternsindata;–Thisbookisabout——Techniquesforfindinganddescribingstructuralpatternsindata.–structuralpatterns表示法：表、树、规则概念：MachineLearningTolearn:–togetknowledgeofstudy,experience,orbeingtaught;–tobecomeawarebyinformationorfromobservation;–tocommittomemory;–tobeinformedof,ascertain(确定);toreceiveinstructionShortcomingswhenitcomestotalkingaboutcomputes–It’svirtuallyimpossibletotestiflearningasbeanachievedornot.–Thistieslearningtoperformanceratherthanknowledge简单例子：天气问题*天气数据：weather.nominal.arff运行Weka，载入数据，选择算法id3预测（决策树）•outlook=rainy•|windy=TRUE:no•|windy=FALSE:yes测试方法：采用10Cross-validation的测试结果：ConfusionMatrix(P.138)和准确率•ab<--classifiedas•81|a=yes•14|b=no•CorrectlyClassifiedInstances1285.7143%•IncorrectlyClassifiedInstances214.2857%其他算法：NeuralNetwork数据挖掘的过程步骤：见『回顾：DM的步骤』输入：Concepts,Instances,AttributesConcept–四种基本的学习类型•Classification,association,clustering,numericprediction–不考虑类型，我们把要学习的称为Concept，而把学习的输出成为conceptdescriptionInstance：数据样本记录Attribute：数据字段–Nominal：outlook:sunny=>no–Ordinal：距离无法度量，如hot>mild>cool–Interval：距离可度量，如整数–Ratio：如58.1%输入：Preparingtheinput*Gatheringthedatatogether–Thedatamustbeassembled,integrated,andcleanedup（DataWarehousing）–Selectingtherighttypeandlevelofaggregationisusuallycriticalforsuccess属性类型：•ARFF文件格式（备注：weather.nominal.arff）•支持两种基本类型：nominalandnumeric，尽可能用前者属性值•Missingvalue：去掉该样本、替代、（用?来表示字段值）•Inaccuratevalue：一粒老鼠屎——需要领域知识！Gettingtoknowyourdata!•数据清理一个耗时、费力，却很重要的过程，•Garbagein,garbageout!输出：KnowledgerepresentationDecisiontablesDecisiontreesClassificationrules•IfaandbthenxAssociationrules：多个结果•If…thenoutlook=sunnyandhumidity=highRuleswithexceptions(P.66)•If…then…except…else…except…TreesfornumericpredictionInstance-basedrepresentationClusters算法：ThebasicmethodsSimplicity-first：simpleideas...