6252

5840

كارشناسي ارشد

نرم افزار

اصفهان: دانشگاه صنعتي اصفهان،دانشكده برق و كامپيوتر

1390

نه،98ص.: مصور،جدول،نمودار

ص.ع.به فارسي و انگليسي

محمد حسين سرايي، پژمان خديوي

10/7/90

محمد داورپناه جزي، رسول موسوي

مهندسي برق و كامپيوتر

ID5840

به فارسي و انگليسي: قابل رويت در نسخه ديجيتالي

99 A Hybrid Data Mining Approach to Credit Scoring with Application to the Bank Customers Ali shalbafzadeh a shalbafzadeh@ec iut ac ir 1390 02 03 Department of Ali shalbafzadeh Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiMohammad hossein Saraee saraee@cc iut ac irPejamn Khadivi pkhadivi@ec iut ac irAbstractImprovements in data usage and data collection with help of computer and also use of web and internet asword wide information system make data mining as a good tool in order to help solve problems In the lastyears credit scoring has been changed to a critical job for bank and insurance organization In fact wrongmanagement in credit assignment area makes a great loss to financial institution and even bankruptcy in theU S A and Europe and even makes them close to bankruptcy Credit scoring focused on decrease the risk ofcredit allocation This task has been done with correct prediction of customers behavior In fact creditscoring problem can be defined as classification problem which targets create useful model that classifycustomers into good and bad groups In the past classics methods like linear regression has been used in orderto create credit scoring model but recently researchers focused on artificial intelligence and machine learningtools to implement different single or hybrid method to solve this problem in order to and help withperformance and optimization model The purpose of this thesis is to develope integrated model with highaccuracy and precision for credit scoring To achieve this model Rough set theory has been used to reduceattributes linear discretization and K Means has been used for discretization and Genetic algorithm andBayes network has been used for classification firstly there are two options for discretization to use Kmeansor linear secondly two option reduce attributes wheter to use rough set or not finally there are four optionsmade up from combination of Bayesian network and genetic Totally they make up 16 different models Theproposed models have applied on Australian German Japaneese and Iranian datasets The results show thatbest proposed model for each dataset produced better than existing approaches The thesis is cocluded bysuggesting areas of future worksKeywordsCredit scoring KMeans Genetic Na ve Bayes Bank Roughset Data mining

محمد حسين سرايي، پژمان خديوي

محمد داورپناه جزي، رسول موسوي