Publication: Tıbbi karar destek sisteminin veri madenciliği yöntemleriyle gerçekleştirilmesi
Abstract
TIBBİ KARAR DESTEK SİSTEMİNİN VERİ MADENCİLİĞİ YÖNTEMLERİYLE GERÇEKLEŞTİRİLMESİ Tıp alanında teşhis ve tedavi kayıtlarının bilgisayar programları tarafından analiz edilmesi etkili tedaviyi destekleyici bir unsur oluşturmaktadır. Birçok alanda olduğu gibi, tıp alanında da veri madenciliği yöntemlerinin kullanımı hızla artmaktadır. Veri madenciliği yöntemlerinin kullanıldığı tıbbi karar destek sistemi tasarımı gerçekleştirerek tıbbi veri setlerinden elde edilecek sonuç ile hekimlere destekleyici yardımda bulunacak sistemin geliştirilmesi bu tezin başlıca amacıdır. Geliştirilen sistemde iki ayrı veri setinde çalışma yapılmıştır. Birinci veri setinden metastatik kolorektal kanser hastalığının hayatta kalma durumu ve ikinci veri setinden ise diyabet hastalığı veri madenciliği yöntemleri kullanılarak tahmin edilmiştir. Çalışmada, sınıflandırma tahmin modeli ve hibrid tahmin modeli olmak üzere iki ayrı tıbbi karar destek sistemi modeli geliştirilmiş ve iki ayrı modelden alınan sonuçlar incelenerek karşılaştırılmıştır. Veri madenciliği sürecindeki veri ön işleme aşamasından sonra, verilerin sınıflandırılması aşamasında, Destek Vektör Makineleri, Naive Bayes, Karar Ağaçları, Yapay Sinir Ağları, Çok Katmanlı Algılayıcı, Lojistik Regresyon algoritmaları kullanılmıştır Metastatik kolorektal kanser veri setinde, yalnızca sınıflandırma algoritmalarının uygulandığı sınıflandırma tahmin modelinde, en başarılı algoritma Destek Vektör Makineleri iken Hibrid tahmin modelinde, karar ağaçları ve yapay sinir ağları %100 doğruluk oranı ile en başarılı algoritmalar olmuşlardır. Sınıflandırma işlemleri sonucunda modellerin doğruluk değerlerine bakıldığında, Sınıflandırma tahmin modelinin doğruluk oranı %65-70 iken Hibrid tahmin modelinde bu oranın %95-100’lere ulaştığı görülmektedir. Gerçekleştirilen hibrid yapı ile yani kümeleme ile sınıflandırma algoritmalarının birleştirerek uygulanmasında doğruluk oranlarının tüm algoritmalar için çok yüksek ve birbirine çok yakın değerler çıkardığı görülmüştür. Diyabet hastalığı analizinde, önişlemden geçirilen diyabet verileri K-means kümeleme algoritmasıyla kümelenmiş ve kümeleme sonucu sınıfları yanlış tahmin edilen veriler veri setinden silinmiştir. Yeni veri seti üzerinde seçilen veri madenciliği sınıflandırma algoritmaları denenmiş ve algoritma başarımlarının değerlendirilmesi, Weka ve RapidMiner veri madenciliği programlarıyla gerçekleştirilmiştir. İki programın da tüm algoritmalar için çok yüksek başarı oranlarına sahip olduğu görülmüştür. Karşılaştırma algoritmalar açısından yapıldığında Çok Katmanlı Algılayıcı algoritması iki programda da en yüksek başarı yüzdesi ile en iyi algoritma, karar ağaçları algoritması ise iki program için de başarı yüzdesi en düşük algoritma olmuştur. Bulgularımız geliştirilen sistemin kullanılmasının, hekimlere karar verme aşamasında yardımcı olacağını göstermektedir.
VERIFYING MEDICAL DECISION SUPPORT SYSTEM WITH THE METHODS OF DATA MINING The analysis of the diagnosis and the treatment records in the field of Medicine by computer programs forms a supportive element in efficient treatment. As in many fields, the use of the methods of data mining has been increasing rapidly in Medicine. Developing the system which can help the doctors is the main purpose of this thesis with designing the medical decision support system and getting the results from medical data sets. In the developed system, studies have been carried out in two different data sets. In the first data set, survival of metastatic colorectale cancer disease, and in the second data set, survival of the disease of diabetes have been estimated with the application of data mining. In the study, two medical decision support systems; classification prediction model and hybrid prediction model have been developed, and the results from both models have been compared by analysing. In the process of data mining, after the stage of pre-processing of data, in the stage of classifying the data, Support Vector Machine, Naive Bayes, Decision Trees, Artificial Neural Networks, Multilayer Perceptron and Logistic Regression algorithms have been applied. In the prediction model in which only classification algorithms have been used, in the data set of metastatic colorectale cancer, the most successful algorithm has been Support Vector Machine, but in Hybrid Prediction Model, decision trees and artificial neural networks have been the most successful algorithms with the rate of a hundred percent accuracy. When the accuracy rate of the models are examined at the end of the classifying process, the accuracy rate of classification prediction model has been %65-70, the accuracy rate of hybrid prediction model has reached %95-100. It has been observed that the accuracy rates are very high and very close to each other for all algorithms in the applicaiton of combining hybrid structure, which can be called clustering with classification algorithms. In the analysis of diabetes disease, the diabetes datas, which were processed pre- processing, have been clustered with the K-means clustering algorithm and at the end of the clustering, the datas whose classes have been estimated wrongly have been erased from the data set. Data mining classifying algorithms have been experimented on a new data set and the evaluation of the success of the algorithms has been made with Weka and RapidMiner data mining programs. Two programs have been observed to have very high success rates for all algorithms. When the comparison made in terms of algoritms, Multilayer Perceptron algorithm is the best algorithm with the highest success rate in both programs and decision trees has the lowest success rate for both programs. Our findings indicates that the usage of the developed system can help the doctors in the decision making process.
VERIFYING MEDICAL DECISION SUPPORT SYSTEM WITH THE METHODS OF DATA MINING The analysis of the diagnosis and the treatment records in the field of Medicine by computer programs forms a supportive element in efficient treatment. As in many fields, the use of the methods of data mining has been increasing rapidly in Medicine. Developing the system which can help the doctors is the main purpose of this thesis with designing the medical decision support system and getting the results from medical data sets. In the developed system, studies have been carried out in two different data sets. In the first data set, survival of metastatic colorectale cancer disease, and in the second data set, survival of the disease of diabetes have been estimated with the application of data mining. In the study, two medical decision support systems; classification prediction model and hybrid prediction model have been developed, and the results from both models have been compared by analysing. In the process of data mining, after the stage of pre-processing of data, in the stage of classifying the data, Support Vector Machine, Naive Bayes, Decision Trees, Artificial Neural Networks, Multilayer Perceptron and Logistic Regression algorithms have been applied. In the prediction model in which only classification algorithms have been used, in the data set of metastatic colorectale cancer, the most successful algorithm has been Support Vector Machine, but in Hybrid Prediction Model, decision trees and artificial neural networks have been the most successful algorithms with the rate of a hundred percent accuracy. When the accuracy rate of the models are examined at the end of the classifying process, the accuracy rate of classification prediction model has been %65-70, the accuracy rate of hybrid prediction model has reached %95-100. It has been observed that the accuracy rates are very high and very close to each other for all algorithms in the applicaiton of combining hybrid structure, which can be called clustering with classification algorithms. In the analysis of diabetes disease, the diabetes datas, which were processed pre- processing, have been clustered with the K-means clustering algorithm and at the end of the clustering, the datas whose classes have been estimated wrongly have been erased from the data set. Data mining classifying algorithms have been experimented on a new data set and the evaluation of the success of the algorithms has been made with Weka and RapidMiner data mining programs. Two programs have been observed to have very high success rates for all algorithms. When the comparison made in terms of algoritms, Multilayer Perceptron algorithm is the best algorithm with the highest success rate in both programs and decision trees has the lowest success rate for both programs. Our findings indicates that the usage of the developed system can help the doctors in the decision making process.