Bobrek nakli gecirmis hastalarda akilli yuntem tabanli yeni oznitelik secme algoritmasi gelistirilmesi
Loading...
Date
2012
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Kadir Has Üniversitesi
Abstract
Veri madenciligi verilerden kesfedilecek desenler yardimiyla yeni bilgiler elde etme amaciyla cok farkli disiplinlerde kullanilan cesitli metotlardan olusmaktadir. Tip alanindaki verinin buyuklugu ve hayati onem tasimasi Veri madenciliginin bu alanda da uygulanmasini gerekli kilmistir. Bu tezde Veri Madenciliginin Tip alaninda kullanimi incelenmistir. Uygulama calismasi icin Ýstanbul universitesi Cerrahpasa Tip Fakultesi.nde ayakta tedavi goren hastalar arasindan Mart 2006 . Aralik 2007 tarihleri arasinda 21 aylik bir surede tedavisi gormus hastalara ait veriler bir araya getirilerek bir veri kumesi olusturulmustur. Bu veri kumesi uzerinde WEKA yazilimi kullanilarak siniflama kumeleme ve karar agaci algoritmalari calistirilmis elde edilen karar kurallari uzman destegiyle incelenerek koroner arterlerde kalsifikasyon bulunmasinda etkili olan faktorlerin neler oldugu belirlenmis ve oznitelik secme algoritmalariyla ayni faktorlere ulasilip ulasilamadigi belirlenmistir.
Data mining consists up of many different methods which try to find new information from data patterns. This is the main reason why it has been a basis for many research areas. The amount of data which belongs to the field of medicine is extensive and also very significant this is the main reason behind why the usage of data mining on these types of datasets have been needed. In this thesis the usage of data mining on the field of medicine has been investigated. The dataset consists of the data from the outpatients of the University of Istanbul - Cerrahpasa Medical Faculty which were treated throughout the period of 21 months between the dates March 2006 - December 2007. With the aid of the WEKA software the dataset was examined with classification, clustering and decision tree algorithms and some decision rules where found. These decision rules were then analysed with the help of specialists to determine which features caused complications in the coronary arteries. Also a comparison with feature selection algorithms were done to see if the same features could be found.We solve optimization problem by using Harmony Search algorithm, by taking cross validation results as objective function and using Naive Bayesian classification algorithm. Our program gets the patient data prepared in Weka and uses it as input to Matlab, a commercial package developed for performing calculations using matrix operations. Optimization, classification and cross validation modules were programmed in Matlab.
Data mining consists up of many different methods which try to find new information from data patterns. This is the main reason why it has been a basis for many research areas. The amount of data which belongs to the field of medicine is extensive and also very significant this is the main reason behind why the usage of data mining on these types of datasets have been needed. In this thesis the usage of data mining on the field of medicine has been investigated. The dataset consists of the data from the outpatients of the University of Istanbul - Cerrahpasa Medical Faculty which were treated throughout the period of 21 months between the dates March 2006 - December 2007. With the aid of the WEKA software the dataset was examined with classification, clustering and decision tree algorithms and some decision rules where found. These decision rules were then analysed with the help of specialists to determine which features caused complications in the coronary arteries. Also a comparison with feature selection algorithms were done to see if the same features could be found.We solve optimization problem by using Harmony Search algorithm, by taking cross validation results as objective function and using Naive Bayesian classification algorithm. Our program gets the patient data prepared in Weka and uses it as input to Matlab, a commercial package developed for performing calculations using matrix operations. Optimization, classification and cross validation modules were programmed in Matlab.