earticle

영어

The data mining technique is applied in various fields as a method to extract information based on massive data, and Bayesian networks are also utilized as useful modeling technique. Accordingly, many algorithms in Bayesian networks such as K2, TAN in expansion have been proposed, and suitability of algorithm for each situation evaluation stage has been requested based on performance test result validation to selectively use optimum algorithm for certain situation. As massive various that affects the result exists in actual situation, acquired information through certain data mining technique is considerably limited. Also, the filmed medical images may positively affect the diagnosis but due to high weight on subjective judgment, it is an abstruse problem to process with automatic system. Through this, improved expansion model of search algorithm is proposed with the K2 or TAN in Bayesian networks, which is relatively advantageous in handling the complicated situation of reality and is based on multivariate probability model. Now, because of the nature of extended Bayesian network which greatly varies the performance depending on the type of applied search algorithm, realistic evaluation is required on performance and suitability of each techniques. So in this thesis, experimentation by using equivalent data on disease diagnosis in extended Bayesian network is conducted, and measured classification accuracy while giving changes in search algorithm such as K2 and TAN. In the experiment, comparative evaluation of performance is done based on the result analysis of 10-fold cross validation, and made it possible to distinguish high risk data through classifying HRCT images of patients with high risk of reoccurring of the disease.