Fraud detection in virtual worlds

Businesses are encouraged to conduct processes such as hosting annual meetings and prototyping in virtual environments in order to save on operational costs. Unfortunately widespread participation may be curtailed when fraudulent behaviour can be rife in unregulated virtual economies. Conversely, overly-complex fraud-prevention mechanisms deployed to handle voluminous amounts of transactional data may run the risk of alienating virtual worlds from the users the measures were meant to protect. The paper presents the multi-agent based Virtual Transactions Tracking Analysis system designed to run in the background to record the trail of virtual currencies for the purpose of analysing and identifying potentially fraudulent activities in virtual transactions. An implementation of the proposed system is adapted for deployment to monitor the flow of virtual currencies and transactional activities in an OpenSimulator civic virtual environment. The build-up of extensive tracking data is then used to provide the source of training sets from which fraud detection rules relating to monetary-like exchanges may be derived through means of machine learning. The generated rules are then applied by the multi-agent system to demonstrate how taking additional transaction data such as currency movements into consideration when evaluating current virtual transactions can improve the overall success of fraud detection.