Behaviour Mining for Fraud Detection
Jianyun Xu, Andrew H. Sung, Qingzhong Liu · 2009
Despite significant efforts by merchants, card issuers and law enforcement to curb fraud, online fraud continues to plague electronic commerce web sites. More advanced solutions are desired to protect merchants from the constantly evolving problem caused by fraud. The supervised machine learning technique for the most well known fraud detection algorithms makes them inadequate for an online system, which usually contains a mammoth size of non-stationery data. This paper describes a method to dynamically create user profile for the purpose of fraud detection. We use a data mining algorithm to adaptively profile legitimate customer behaviour in a form of association rule set from a transaction database. Then the incoming transactions are compared against the user profile to indicate the anomalies. A novel pattern match approach is proposed to evaluate how unusual the new transactions are. An empirical evaluation shows that we can accurately differentiate the anomaly behaviour from profiled user behaviour.