A new DCT based scalable distributed fraud detection architecture

In the past few years we have experienced a constant increase in the number of fraud cases in electronic transactions. This emphasizes the necessity for an efficient, yet accurate online fraud detection system. The architecture of most current online transaction systems such as electronic banking is distributed in nature. Considering the extremely large number of transactions that is carried out by such distrusted systems, one of the most important and vital characteristics of an efficient fraud detection framework in such an environment, is scalability. In this paper, we propose a scalable distributed fraud detection architecture. The proposed architecture has a hierarchical structure to increase its scalability. The proposed architecture has been evaluated in terms of accuracy and scalability using synthetically generated data as well as benchmarks for network intrusion detection (KDD99). Results indicate that the proposed method has an order of O(1) for the required amount of processing, storage and bandwidth resources in each node, compared to conventional client-server architectures that demonstrate a linear relationship for server with these resources.