Credit Risk Analysis Based on Hybrid Classification: Case Studies on German and Turkish Credit Datasets

Abstract

In finance sector, credit risk analysis plays a major role in decision process. Banks and finance institutions gather large amounts of raw data from their customers. Data mining techniques can be employed to obtain useful information from this raw data. Several data mining techniques, such as support-vector machines (SVM), neural networks, naive-bayes, have already been used to classify customers. In this paper, we propose hybrid classification approaches, which try to combine several classifiers and ensemble learners to boost accuracy on classification results. Furthermore, we compare these approaches' performance with respect to their classification accuracy. We work with two diverse datasets; namely, German credit dataset and Turkish bank dataset. The goal of using such diverse dataset is to show generalization capabality of our approaches. Experimental results provide three important consequences. First, feature selection stage has a major role both on result accuracy and calculation complexity. Second, hybrid approaches have better generalability over single classifiers. Third, using SVM-Radial Basis Function (RBF) as the base classifier and a hybrid model member gives the best accuracy and type-1 accuracy results among others.