Affiliations: [a] Department of Economics, University of Ioannina, Ioannina, Greece | [b] Department of Materials Science and Engineering, University of Ioannina, Ioannina, Greece | [c] Department of Applications of Information Technology in Administration and Economy, Technological Educational Institute of Ionian Islands, Lefkada, Greece
Abstract: Credit scoring or credit risk assessment is a domain of major importance for financial institutions. Accurate predictions can lead to significant savings for the institutions. In the current work we evaluate the use of an Ant Colony System (ACS) in the problem of credit scoring. ACS are nature inspired algorithms that search for the optimal solution, by mimicking the functions of ants. In our application artificial ants are applied for rule extraction. The performance of ant based rule extraction is compared against six widely used classification methods. All tests are complemented with feature selection approaches, for dimensionality reduction. Our evaluation is performed using three different datasets with credit scoring instances. The obtained results indicate that the examined ant based approach, offers high accuracy comparative to the accuracies obtained by the rest of the classifiers. Considering the fact that our approach has the ability to extract classification rules, thus offering interpretation of results, it appears as a promising alternative classification method for credit scoring.
