7463

6969

كارشناسي ارشد

نرم افزار

اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر

1391

نه، 93ص.: مصور، جدول، نمودار

ص.ع.به فارسي و انگليسي

محمدعلي منتظري، محمد صنيعي آباده

25/10/91

رسول موسوي، عبدالرضا ميرزايي

1396/09/22

كتابنامه

برق و كامپيوتر

مهندسي برق و كامپيوتر

ID6969

به فارسي و انگليسي: قابل رويت در نسخه ديجيتالي

Designing and Implementation an Ensemble Algorithm Based On Decision Trees for Customer Credit Scoring Hamid Saberi h saberi@ec iut ac ir Date of Submission Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisors Mohammad Ali Montazeri Montazeri@cc iut ac ir Mohammad Saniee Abade Saniee@Modares ac ir Abstract Credit scoring is a method that banks and financial institutions employ it with the current and past information of applicants to evaluate the probability of not reimbursement of loans and also to grant them scores Credit scoring models generally categorize credit applicants based on finanicial factors in to two classes the good credit class that is able to perform financial commitments and the bad credit class that should not be granted credit due to the high probability of defaulting on the financial commitments Credit scoring is an analytical technique for risk assessment Credit risk is the most challenging risk to which financial institution are exposed The huge amounts of waste or deferment loans indicate the lack of suitable models and systems to evalute and manage credit risks Credit scoring system is one ot the main tools to manage and control credit risks Regarding to the enoumous growth of information and experiences in banking industry specially in two decades ago and also the growth of potentials credit applicants it needs to develop more completed and sophisticated models that can automatically perform credit granting and supervise people finanical health Since an improvement in accuracy even as a small percent might led into significant savings more sophisticated models should be proposed for significantly improving the accuracy of the credit scoring models In this thesis we have proposed a hybrid credit scoring model based on Adaboost and Decision Trees Algorithm DTA In this model several decision trees are aggregated and formed a powerfull classifier This process is implemented sequentially Adaboost assignes a coefficient to each tree base on classification accuracy of the tree and also improvement of previous trees deficiencies on samples classifications The final classifier classifies samples based on these assigned coefficients In fact the main concentration of this algorithm is on selecting more accurated trees This property improves accuracy stability and reduces overfitting problems The capability of this hybrid method is evaluated using basic performance measurements e g accuracy sensitivity and specificity Receiver Operating Characteristic ROC curve and Area Under Curve AUC score Two real data sets have been used To demonestrate our model we have developed a software based on c# language and SQL server as a deta base engine All of the performance measures are also implemented in this software This software capable of creating thousand decision trees on the inpute data set Experimental results indicates that the proposed method outperforms a single classifier and other combined classifiers for the credit scoring prediction in terms of accuracy and efficiency

محمدعلي منتظري، محمد صنيعي آباده

رسول موسوي، عبدالرضا ميرزايي