Procurement Card Fraud Detection Using Hidden Markov Models and Information Fusion: A Fusion

• In Event: SET Section

Abstract

Internal Anomaly Detection in the area of employee procurement cards continues to be an issue within many firms, even large multinational corporations. In this work in progress study, the authors observed the manual procedures that were being employed by a firm and created detailed scripts imitating and enhancing these procedures, all in the process of developing an expert system. However, many transactions are missing important information such as items purchased and these transactions were ignored by the firm. Furthermore, the system could not dynamically learn new instances of personal use that had not been previously identified and classified. In this comparative study, two approaches are utilized to analyze these transactions: Hidden Markov Models and Information Fusion.
In this procurement card study, the details of the items purchased in individual transactions are not known to the issuing bank and to the firm. This unknown information is then represented as the underlying finite Markov chain, which is NOT observable. The authors will model a procurement card processing sequence using this hidden random process of an HMM.
Information Fusion is a hybrid model of Bayesian Belief Networks and Dempster-Shafer Theory. The goal of this system will be to develop a reasonably accurate anomaly detection system that can dynamically learn the behavior of the procurement card users over time, and thus minimize losses to the firm.
This research is important as it contributes to the literature of anomaly detection by an expert system of procurement card transactions. By applying the supervised knowledge that was acquired from the development of the iLisa tool and from the fusion of these two approaches, HMM and Information Fusion, the authors hope to contribute to the development of a self-learning, omniscient expert systems tool.

Authors

• Deniz A Appelbaum, Rutgers, The State University of New Jersey, Newark

• Abdullah Alawadhi, Rutgers, The State University of New Jersey