University of Piraeus (UNIPI)
Εφαρμογές πληροφορικής στην αξιολόγηση μεθόδων κατασκευής υποδειγμάτων εκτίμησης της πιστοληπτικής ικανότητας των υποψηφίων πιστούχων στον τομέα της λιανικής τραπεζικής
Abstract
dc:descriptionOne of the most important problems in the retail banking sector is that of evaluating the creditworthiness of credit applicants. This problem is addressed by the methodology of credit scoring. The methodology of credit scoring involves the construction of models (i.e. credit scoring models) which classify each credit applicant into one of two classes: the class of applicants who are predicted not to default with their obligations and are granted credit and the class of applicants who are predicted to default with their obligations and are not granted credit. The goal of this thesis is to study the methodology of credit scoring and how this methodology can be implemented by means of computer software applications. In pursuing this goal, this thesis examines the theoretical background of the methodology of credit scoring and analyzes credit scoring data by means of the XLMiner software package and constructed software templates. Within the framework of credit scoring data analysis, methods used for the construction of credit scoring models are assessed. Particular emphasis is given into the method known as the Naive (Simple) Bayes Rule (NBR). Specifically, a real-world credit scoring data set from a major Greek bank is used to benchmark NBR against linear discriminant analysis, logistic regression, neural networks, classification trees and k-nearest neighbours. Five partitions of this data set into training and test sets are used. Predictive ability is measured by the percentage of correctly classified cases and the bad rate amongst accepts. NBR appears to have a lower predictive ability than the other methods under both the aforementioned predictive ability measures. This result may be attributed to properties inherent in the field of credit scoring, i.e. significant dependence given the class of the variables used in this field and large training set sizes. The results allow the application of NBR as a method for the construction of credit scoring models, because of the small (statistically non-significant) differences in predictive ability between the methods and the simplicity of NBR.
Degree
thesis:*- Grantor dc:publisher
- University of Piraeus (UNIPI)
- Year dc:date
- 2009
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Αντωνάκης, Αντώνιος
- Antonakis, Antonios
Subjects
dc:subject × 20- Εκτίμηση πιστοληπτικής ικανότητας
- Ταξινόμηση
- Κανόνας απλού (αφελούς) Bayes
- Πιστωτικός κίνδυνος
- Λιανική τραπεζική
- Διαχωριστική (διακριτική) ανάλυση
- Λογαριθμική (λογιστική) παλινδρόμηση
- Δίκτυα, Νευρωνικά
- Credit scoring
- Classification
- Simple (Naive) Bayes Rule
- Credit risk
- Retail banking
- Discriminant analysis
- Logistic regression
- Networks, Neural
- Κοινωνικές Επιστήμες
- Οικονομικά και Επιχειρήσεις
- Social Sciences
- Economics and Business
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/18638