Multi-agent system (MAS) support for credit referencing
Ndirangu, Peter A
MetadataShow full item record
With the ever increasing demand for credit facilities, mechanisms and models for early warning have been put in place whereby financial institutions share information regarding their customers‟ credit behavior with the establishment of credit reference bureaus (CRBs). Banks and other lending financial institutions not only form CRBs' main clientele, but also are the data furnishers. However, most of these data furnishers do not have a fully functional integrated CRB solution thereby making extraction, validation and preparation of CRB data quite a task; leading to inaccurate credit reports produced from CRBs‟ data warehouses that usually are not up-to-date. Agent-based systems technology presents an opportunity to make credit risk decision making timelier, efficient and less human centric. In this project, we develop a Multi-Agent System (MAS) supported credit referencing platform with the capability to automatically integrate customer credit information data from the heterogeneous and geographically distributed archives of the data furnishers to the CRB‟s data warehouse, perform credit risk evaluation and generate credit reference reports in real-time. The platform encompasses uses a task agent to coordinate communication between the web user interface and the tools. The data integration tool uses a data collection agent that updates the CRB data warehouse with data received from the external database wrapper agents. The credit risk evaluation tool uses a data analysis agent to retrieve customers‟ credit information from the CRB‟s data warehouse and evaluate the risk levels posed by credit applicants over a set of pre-defined criteria defined by the Fair Isaac Corporation (FICO). For testing the prototype, customer data from four banks is integrated and customer credit risk levels evaluated. For evaluation, a simple match of prototype approved datasets to human approved datasets on twenty real-world credit applicants is done and 95% accuracy attained.