Zahra Davoodabady - Computer Eng. Department, Shahab-e-Danesh Institute of Higher Education, Qom, Iran
Ali Moeini - Algorithms and Computations Department, University of Tehran, Tehran, Iran

Today's financial transactions have increased with banks and financial institutions. Attempts to find a credit scoring model with high accuracy has become a competition between financial institutions. We have created 9 different models for the credit scoring of customers by combining three methods of feature selection and three decision tree methods. The model is implemented on three dataset and we compare the accuracy of the models. The two datasets choose from the UCI (Australian dataset, German dataset) and a given dataset is a car leasing company in Iran. In this paper we combine ReliefF algorithm as feature selection methods with decision tree learning algorithm ID3, C45 and CART. The proposed methods is described and compared based on classification accuracy and type I and II error rate. Results compare with classification models without feature selection algorithm, too. Results show that using feature selection methods with decision tree algorithm build more accurate models.

credit scoring, data mining, decision tree, feature selection, ReliefF