An agent-based service oriented architecture for risk mining

Publication Type:
Issue Date:
Full metadata record
Risk Mining (RM) is the process of analyzing data including risk information by data mining methods, with the mining results for risk prevention. In the last few years, some researchers have proposed the combination of data mining and agent technology (agent mining) to improve the performance of data mining methodology in the heterogeneous business environments. However, problems exist for further research with the application of risk mining systems in real industry environments to enhance the robustness of system architect, dynamic business process and model accuracy etc. Therefore, in this thesis we present an Agent-based Service-oriented Risk Mining Architecture (ABSORM), which has been designed to facilitate the development of agent mining systems to address the above issues. This thesis focuses on developing the following strategies: • The integration of agent technology with web service. In this framework, we propose a new and easier method, by which the system functions are not integrated into the structure of the agents, rather modeled as distributed services and applications which are invoked by the agents acting as controllers and coordinators. Therefore, techniques developed in this framework can improve the interoperability between different modules, distribution of resources, and the lack of dependency of programming languages. • The integration of agent technology with business process management. In this work, we develop the autonomous agents that can collaborate in a business flow, which not only increases the reusability of the system, but also eases the system development in terms of re-usability of the computational resources. A group of agents solves problems in the following way: each individual agent solves the problem individually, and then interacts with each other to finalize a business process. • The integration of agent technology with ensemble learning methods. In this thesis, we are interested in developing agent-based ensemble learning strategies for risk mining: each ensemble agent individually gathers the evidence about model evaluation, and then ensembles learning methods like bagging and boosting is used to obtain prediction from the individually gathered evidence. Agent based ensemble learning can provide a critical boost to risk mining where predictive accuracy is more vital than model interpretability. The proposed architecture has been evaluated for building an online banking fraud detection system and a student risk management system. These two applications have been proved to be a sophisticated, yet user friendly, risk analysis and management tool. They are modular, interactive, dynamic and globally oriented.
Please use this identifier to cite or link to this item: