基于代价敏感主动学习算法的2型糖尿病诊断

doi:10.3969/j.issn.1006-2475.2018.06.016

摘要/Abstract

摘要： 建立2型糖尿病诊断模型，并通过主动学习解决医疗数据中标记样本较少的问题。2型糖尿病的诊断可以被看作一个代价敏感的二分类问题，本文基于逻辑回归模型、支持向量机模型和人工神经网络模型，采用基于期望误差减小的代价敏感主动学习方法，将主动学习算法和代价敏感分类算法相结合来构建诊断模型，将不同的误分类代价考虑到样本的选择中。在2型糖尿病诊断问题中，基于期望误差减小的代价敏感主动学习算法表现最优，以较少的样本标记达到了最低的误分类代价，因此主动学习算法能够减少医疗数据挖掘中需要标记的样本数，节省标注成本，同时保证模型的性能。

关键词: 糖尿病, 诊断模型, 代价敏感分类, 主动学习, 逻辑回归, 支持向量机, 人工神经网络

Abstract: Inthisstudy,adiagnosismodelfortype2diabeteswasbuiltandthelabelabsenceprobleminmedicaldatawassolvedbyactivelearning.Thediagnosisoftype2diabetescanbeseenasacost-sensitivebinaryclassificationtask.Takinglogisticregression,supportvectormachines(SVM)andartificialneuralnetwork(ANN)asthebasemodel,thisstudyadoptedthecost-sensitiveactivelearningalgorithmbasedontheexpectederrorreductionframework,whichcombinedtheactivelearningstrategywiththecost-sensitiveclassificationalgorithmandintroducedthecostinformationintotheinstancesamplingprocess.Forthediagnosisoftype2diabetes,thecost-sensitiveactivelearningalgorithmbasedontheexpectederrorreductionframeworkperformedbestinthesecomparedactivelearningstrategiesanditachievedtheminimummisclassificationcostsbylabelingfewerinstances.Activelearningalgorithmscanreducethenumberofinstancestobelabeled,savethelabelingcostsandguaranteethemodelperformanceatthesametime.

Key words: diabetes；diagnosticmodel；cost-sensitiveclassification；activelearning, logisticregression；supportvectormachine；artificialneuralnetwork

中图分类号:

TP181

许智彪. 基于代价敏感主动学习算法的2型糖尿病诊断[J]. 计算机与现代化, doi: 10.3969/j.issn.1006-2475.2018.06.016.

XUZhi-biao. DiagnosisofType2DiabetesBasedonCost-sensitiveActiveLearningAlgorithm[J]. Computer and Modernization, doi: 10.3969/j.issn.1006-2475.2018.06.016.

参考文献［22］

［1］	NathanDM,BuseJB,DavidsonMB,etal.Medicalmanagementofhyperglycemiaintype2diabetes:Aconsensusalgorithmfortheinitiationandadjustmentoftherapy［J］.ClinicalDiabetes,2009,27(1):4-16.
［2］	WhitingDR,GuariguataL,WeilC,etal.IDFdiabetesatlas:Globalestimatesoftheprevalenceofdiabetesfor2011and2030［J］.DiabetesResearchandClinicalPractice,2011,94(3):311-321.
［3］	NorrisSL,KansagaraD,BougatsosC,etal.Screeningadultsfortype2diabetes:AreviewoftheevidencefortheU.S.PreventiveServicesTaskForce［J］.AnnalsofInternalMedicine,2008,148(11):855-868.
［4］	DetranoR,JanosiA,SteinbrunnW,etal.Internationalapplicationofanewprobabilityalgorithmforthediagnosisofcoronaryarterydisease［J］.TheAmericanJournalofCardiology,1989,64(5):304-310.
［5］	GamboaALG,MendozaMG,OrozcoREI,etal.Hybridfuzzy-SVclusteringforheartdiseaseidentification［C］//IEEEInternationalConferenceonComputationalIntelligenceforModeling,ControlandAutomationandInternationalConferenceonIntelligentAgents,WebTechnologiesandInternetCommerce.2006:121.
［6］	KahramanliH,AllahverdiN.Extractingrulesforclassificationproblems:AISbasedapproach［J］.ExpertSystemswithApplications,2009,36(7):10494-10502.
［7］	CascioD,FauciF,MagroR,etal.Mammogramsegmentationbycontoursearchingandmasslesionsclassificationwithneuralnetwork［J］.IEEETransactionsonNuclearScience,2006,53(5):2827-2833.
［8］	SettlesB.ActiveLearningLiteratureSurvey［R］.UniversityofWisconsin-Madison,ComputerScienceReport1648.2009.
［9］	〖JP2〗ZhangYexun,WangYanfeng,CaiWenbin,etal.Fromtheorytopractice:Efficientactivecost-sensitiveclassificationwithexpectederrorreduction［C］//Proceedingsofthe2017SIAMInternationalConferenceonDataMining.SocietyforIndustrialandAppliedMathematics.2017:153-161.〖JP〗
［10］	LiuA,JunG,GhoshJ.Aself-trainingapproachtocostsensitiveuncertaintysampling［J］.MachineLearning,2009,76(2-3):257-270.
［11］	LewisDD,GaleWA.Asequentialalgorithmfortrainingtextclassifiers［C］//Proceedingsofthe17thAnnualInternationalACMSIGIRConferenceonResearchandDevelopmentinInformationRetrieval.1994:3-12.
［12］	ScheinAI,UngarLH.Activelearningforlogisticregression:Anevaluation［J］.MachineLearning,2007,68(3):235-265.
［13］	LongBo,ChapelleO,ZhangYa,etal.Activelearningforrankingthroughexpectedlossoptimization［J］.IEEETransactionsonKnowledgeandDataEngineering,2015,27(5):1180-1191.
［14］	LopezV,FernandezA,Moreno-TorresJG,etal.Analysisofpreprocessingvs.cost-sensitivelearningforimbalancedclassification.Openproblemsonintrinsicdatacharacteristics［J］.ExpertSystemswithApplications,2012,39(7):6585-6608.
［15］	HastieT,TibshiraniR,FriedmanJ.TheElementsofStatisticalLearning:DataMining,InferenceandPrediction［M］.2nded.Springer,2009.
［16］	BaxtWG.Useofanartificialneuralnetworkforthediagnosisofmyocardialinfarction［J］.AnnalsofInternalMedicine,1991,115(11):843-848.
［17］	〖JP2〗RoyN,MccallumA.Towardoptimalactivelearningthroughsamplingestimationoferrorreduction［C］//Proceedingsofthe8thInternationalConferenceonMachineLearning.2001:441-448.〖JP〗
［18］	ElkanC.Thefoundationsofcost-sensitivelearning［C］//InternationalJointConferenceonArtificialIntelligence,2001.2001:973-978.
［19］	CaiWenbin,ZhangYa,ZhouSiyuan,etal.Activelearningforsupportvectormachineswithmaximummodelchange［C］//JointEuropeanConferenceonMachineLearningandKnowledgeDiscoveryinDatabases.2014:211-226.
［20］	ZhouSiyuan,ZhangYa.Activelearningforcost-sensitiveclassificationusinglogisticregressionmodel［C］//IEEEInternationalConferenceonBigDataAnalysis.2016:1-4.
［21］	DomingosP.Metacost:Ageneralmethodformakingclassifierscost-sensitive［C］//Proceedingsofthe5thACMSIGKDDInternationalConferenceonKnowledgeDiscoveryandDataMining.1999:155-164.
［22］	WangTao,QinZhenxing,ZhangShicao,etal.Cost-sensitiveclassificationwithinadequatelabeleddata［J］.InformationSystems,2012,37(5):508-516.

[1]	杨博, 庄毅. 基于AOA-MSVM的控制集群故障检测方法[J]. 计算机与现代化, 2023, 0(12): 112-116.
[2]	刘静乐, 罗翔, 宫成荣, 张国鹏. 基于RF-RFECV和LightGBM算法的糖尿病预测[J]. 计算机与现代化, 2023, 0(11): 36-43.
[3]	杨孙哲, 孙爱珍. 基于HSV颜色与LBP纹理特征的水稻氮素营养诊断[J]. 计算机与现代化, 2023, 0(07): 86-92.
[4]	申志, 李元 . 基于KPCA和SSA优化SVM的非线性过程故障检测#br#[J]. 计算机与现代化, 2023, 0(06): 15-20.
[5]	尹建丰, 卫鑫, 顾雄伟, 黄凯, 魏敏捷. 基于图像阈值优化及改进SVM的电表数字识别[J]. 计算机与现代化, 2023, 0(05): 106-110.
[6]	汪敏, 徐英豪, 朱习军. 基于改进遗传算法优化BP神经网络的糖尿病并发症预测模型#br#[J]. 计算机与现代化, 2022, 0(11): 69-74.
[7]	赵延平, 王芳, 夏杨. 基于支持向量机的短文本分类方法[J]. 计算机与现代化, 2022, 0(02): 92-96.
[8]	肖宏宇, 曾文驱, 王淑营. 基于模型特征匹配的BIM模型混合推荐算法[J]. 计算机与现代化, 2022, 0(01): 28-32.
[9]	刘亚南, 郭南, 赵阳, 余贶琭, . 基于WOA优化神经网络的BOTDA传感信息提取[J]. 计算机与现代化, 2021, 0(12): 19-26.
[10]	赵聪慧, 冯庆胜. 基于DT-CWT和SVM的踏面旋转不变纹理提取算法[J]. 计算机与现代化, 2021, 0(12): 85-90.
[11]	屈晓渊, 张永恒. 基于主动学习的遥感图像地块目标分类[J]. 计算机与现代化, 2021, 0(11): 50-55.
[12]	张岑芳. 基于主动学习的命名实体识别算法[J]. 计算机与现代化, 2021, 0(07): 18-22.
[13]	陈翔, 邹庆年, 谢绍宇, 陈翠琼. 基于迁移学习策略的压板开关状态识别[J]. 计算机与现代化, 2021, 0(05): 120-126.
[14]	李梓龙, 吕勇, 谭国平, 严勤, . 基于纹理特征的超声图像乳腺肿块识别[J]. 计算机与现代化, 2021, 0(02): 1-6.
[15]	王文蔚, 肖军弼, 程鹏, 张悦. 基于SDN的DDoS攻击防御系统[J]. 计算机与现代化, 2021, 0(02): 117-121.