資源描述：
《分類技術(shù)在醫(yī)學(xué)診斷中的應(yīng)用研究》由會(huì)員上傳分享，免費(fèi)在線閱讀，更多相關(guān)內(nèi)容在工程資料-天天文庫。
1、汕頭大學(xué)碩士學(xué)位論文分類技術(shù)在醫(yī)學(xué)診斷中的應(yīng)用研究姓名：楊書鋒申請學(xué)位級別：碩士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：于津20090530摘要癌癥、糖尿病、SARS等重大突發(fā)疾病的早期發(fā)現(xiàn)和確診是疾病成功治療的關(guān)鍵。a前，對這些疾病的診斷主要依靠醫(yī)生的臨床經(jīng)驗(yàn)。論文利用數(shù)據(jù)挖掘的分類功能分析過往臨床數(shù)據(jù)，將醫(yī)生診斷經(jīng)驗(yàn)形式化、客觀化，以提高診斷的準(zhǔn)確率。論文討論了醫(yī)學(xué)數(shù)據(jù)挖掘面臨的問題，綜述了國內(nèi)外在應(yīng)用數(shù)據(jù)挖掘技術(shù)進(jìn)行醫(yī)學(xué)診斷方面的研究現(xiàn)狀和發(fā)展。針對醫(yī)學(xué)數(shù)據(jù)集維度較高、不利于直接處理的問題，論文研究了屬性子集選擇算法，選擇與診斷結(jié)果相關(guān)性較高的屬性子集降以低數(shù)據(jù)維度。在
2、研究分析多種適用于醫(yī)學(xué)診斷的分類方法的基礎(chǔ)上，論文提出基于貝葉斯理論的復(fù)合分類方法(Bayesian-basedCompoundClassificationMethod,BCCM),采用條件概率計(jì)算的方法組合多個(gè)分類器的診斷結(jié)果以提高分類準(zhǔn)確率。運(yùn)用復(fù)合分類器的基本思想，論文對KNN分類算法進(jìn)行改進(jìn)，提出基于貝葉斯理論的模糊KNN分類方法(Bayesian-basedFuzzyKNNClassificationMethod,BFKCM),將每個(gè)最近鄰居看做簡單的分類器，采用基于貝葉斯理論的概率計(jì)算方法組合K個(gè)最近鄰居的預(yù)測結(jié)果以提高分類準(zhǔn)確率。在威斯康辛州乳腺癌數(shù)據(jù)
3、集和比馬印第安人糖尿病數(shù)據(jù)集上進(jìn)行的實(shí)驗(yàn)表明，BCCM和BFKCM在一定程度上提高了疾病診斷的準(zhǔn)確率，有效降低了漏診率。關(guān)鍵詞：醫(yī)學(xué)診斷；屬性子集選擇；貝葉斯理論；BCCM；BFKCMABSTRACTEarlydetectionandaccuratediagnosisareveryimportanttotreatcancer,diabetes,SARSandotherseriousdiseases.Atpresent,itismainlyreliedondoctors?clinicalexperiencetodiagnosethesediseases.Forimpr
4、ovingdiagnosisaccuracy,thispaperattemptstoformalizeandobjectifysuchexperiencesbyutilizingclassificationfunctionofdataminingtechniquestoanalyzinghistoryclinicaldata.Thispaperdiscussestheproblemsaboutminingmedicaldataandgivesanoverviewofresearchinapplyingdataminingtechniquestomedicaldiag
5、nosis?Attributesubsetsselectionmethodsisusedtoselecthighlyrelatedattributesbecauseclinicaldatagenerallyishighdimensionalandcan'tbeanalyzeddirectly?MultipleclassificationmethodsareintroducedandaBayesian-basedCompoundClassificationMethod(BCCM)ispresented,whichmakespredictionbycombiningth
6、epredictionofmultipleclassifiersaccordingtotheiraccuracymatrix.ThesameprinciplebehindBayesiabasedcompoundclassificationmethodisappliedtopresentBayesian-basedFuzzyKNNClassificationMethod(BFKCM),animprovedversionofk-nearestneighborsclassificationmethod,whichconsiderseachnearestneighborof
7、newdatapointtobeasimpleclassifierandpredictstheclasslabelofnewdatapointbyemployingBayesiabasedcompoundclassificationmethod.ExperimentperformedonWisconsinBreastCancerandPimaIndiansDiabetesdatasetsshowsthat,BCCMhasbetterperformancethananyconstructingclassificationmethodandBFKCMhasbette
