An Optimized Type-2 Fuzzy Logic Based System for Detection of Financial Defaults in Sudanese Banking Sector

Abstract

The recent global financial-economic crisis has led to the collapse of several companies from all over the world. This has created the need for powerful models which can predict and reduce potential risks in financial applications. Such models help organizations to enhance the quality and productivity of their services as well as reduce financial risk. The widely used techniques to build predictive models in the financial sector are based on statistical regression, which is deployed in many financial applications such as risk forecasting, customers’ loan default, and fraud detection. However, in the last few years, the use of Artificial Intelligence (AI) techniques has increased in many financial institutions because they can provide powerful predictive models. However, the vast majority of the existing AI techniques employ black box models like Support Vector Machine (SVMs) and Neural Network (NNs) which are not able to give clear and transparent reasoning to explain the extracted decision. However, nowadays transparent reasoning models are highly needed for financial applications. In this thesis, the researcher will present an intelligent Type-2 Fuzzy Logic System for the prediction of financial default to assist the decision-makers in the financial sector to prevent and control the frequent default risk. The Type-2 Fuzzy Logic System would be clear to present a highly explainable and transparent model that is very appropriate to handle different types of uncertainties associated with the financial sector and convert the gathered data to linguistic formats which can be easily stored and analyzed. Fuzzy Logic Based Systems provide a transparent model which provides IF-Then rules that can be easily investigated and understood by the end-users. The proposed model was trained and tested using real data obtained from Albald Bank in Sudan dating back to the period 2007 – 2017. In this study, the two components of Type-1 and Type-2 Fuzzy Logic Systems which are the rule base and fuzzy membership functions are learned dynamically from data. The Fuzzy C-Means (FCM) clustering technique is used to cluster the parameters of the Type-1 and Type-2 Fuzzy Logic controller. To optimize the Type-2 fuzzy logic System the researchers used the Big Bang-Big Crunch (BB-BC) algorithms to optimize the proposed model’s parameters which are the membership functions and the rule base, furthermore, the rule base is optimized in two levels, the number of rules in the rule base and the number of antecedents in the rule itself. As a result of this optimization, the proposed Fuzzy Logic Based System allows achieving relatively high prediction accuracy with optimized membership functions and a small number of rules which increase the System readability and then allow prediction of default risk in the financial sector. The BB-BC optimized type-2 Fuzzy Logic prediction System gained 84% prediction accuracy in our testing dataset which is better than its counterpart’s Type-1 and non-optimized Type-2 Fuzzy Logic prediction system. It is known that the black-box model could not show the details when it provides the prediction, and it is not very good for people to recognize the reasoning behind the given decision, on the contrary, the fuzzy logic prediction system is a white box model which has the transparency that could explain the details behind the reasoning process in giving decision. The results showed that the optimized proposed Type-2 Fuzzy Logic System provided a more interpretable model which has a rational number of rules; only 400 rules which provided a more interpretable rule base that can be easily understood and analyzed by the decision-maker, furthermore the proposed optimized model provided good prediction model which can predict default and serve both sides of business stakeholders; the bank and the customers.