Abstract:
Improved fraud detection systems are vital tools for the evolution of the Sudanese banking
sector where the traditional fraud detection models are incapable of overcoming the emerging, innovative
and new attacks that threaten large financial institutions. Hence, there is a need for accurate and transparent
techniques which can automatically detect fraud with high speed and identify its causes and common
patterns. Many of the Artificial Intelligence (AI) techniques are effective and provide good predictive
models. Nevertheless, they are considered as black-box models. On the other hand, the white box models
are easy to understand and analyze, but result in a large number of rules, besides having many parameters
in each rule. In this paper, we present a novel system based on the Big Bang–Big Crunch optimization (BB–
BC) approach, which is combined with type-2 Fuzzy Logic Systems to result in a small set of short IF-Then
rules for the fraud detection within the Sudanese banking sector. The proposed system uses real-world
dataset from Balad Bank – Sudan, which contains 803,386 transactions with 107 fraud transactions. Hence,
the positive class (frauds) rate is 0.0133% of all transactions. The experimental results demonstrate that the
performance of proposed system is effective in tuning the parameters of the rule base and membership
functions of the Type-2 FLSs (T2FLSs) to improve the accuracy, where the proposed T2FLSs outperformed
the Type-1 FLSs (T1FLSs) counterpart, as well as each rule can be simply explainable. Therefore, this can
be very helpful for the Sudanese banks to start tracking the fraud cases.
