Grant: Collaborative Research and Development Grants

As with many large medical and life insurance business, Medavie Blue Cross has developed a large repository of information from multiple database sources that come from their diversity of clients such as pharmacies, physicians' offices, and clinics, which is continuously being updated with new monthly information for its billing and claims process. Such a wealth of information may be explored to find new patterns of fraud. However this wealth and diversity of data also presents a challenge to this task commonly referred to as the 'curse of dimensionality' coined by Richard E. Bellman related to trying to handle this search in a tractable manner.The objective of this research project is to take advantage of a large repository of information to efficiently search for new fraud patterns across multiple data structures. A new approach for using Markov decision process (MDP) that models audit searches has the ability to handle this flexibility of representation while allowing exploring for fraud. A number of search techniques are possible with a MDP model, which this project explores, to handle the dimensionality growth such as approximate dynamic programming methods of reinforcement learning and linear approximations using computational analysis techniques to maintain precision of actual fraud cases while reducing processing time and storage. The main potential benefit of this study for Medavie Blue Cross and other insurance businesses will be the reduction in costs to the businesses by saving on the loss due to the newly uncovered fraud cases. In addition, the general algorithms and efficiency techniques should be broadly applicable to other Canadian industries that require fraud auditing of large repositories of information beyond the insurance field.