PaySim and RetSim are the two simulators that were developed as part of my research to answer the research questions presented in my PhD thesis. Both simulators are described in detail in the publications.
The Mobile Money Payment Simulation case study is based on a real company that has developed a mobile money implementation that provides mobile phone users with the ability to transfer money between themselves using the phone as a sort of electronic wallet. The task at hand is to develop an approach that detects suspicious activities that are indicative of fraud.
Unfortunately, during the initial part of our research this service was only been running in a demo mode.
This prevented us from collecting any data that could had been used for analysis of possible detection methods.
The development of PaySim covers two phases. During the first phase, we modelled and implemented a MABS that used the schema of the real mobile money service and generated synthetic data following scenarios that were based on predictions of what could be possible when the real system starts operating. During the second phase we got access to transactional financial logs of the system and developed a new version of the simulator which uses aggregated transactional data to generate financial information more alike the original source.
Since we have access to several years’ worth of transaction data from one of the largest Scandinavian retail shoe store chains, we developed RetSim, a Retail shoe store Simulation, built on the concept of MABS. RetSim is designed to be used in developing and testing fraud scenarios at a retail store, while keeping business sensitive and private personal information about customer’s consumption secret from competitors and others. Simulations in the domain of retail stores have traditionally been focused on finding answers to logistics problems such as inventory management, supply management, staff scheduling and for customer queue reductions. To our knowledge, RetSim is the first simulator with the purpose of fraud detection on the retail store domain.
The defense against fraud is an important topic that has seen some study. In the retail store the cost of fraud if of course ultimately transferred to the consumer, and finally impacts the overall economy. Our aim with the research leading to RetSim is to learn the relevant parameters that governs the behaviour in and of a retail store to simulate normal behaviour. However, we also model the malicious behaviour and implement detection techniques. As fraud in the retail setting is usually perpetrated by the staff we have focused on that. Examples of such fraud are explained in
sectionretail and includes: Refunds and Coupon Reductions/Discounts.
In terms of the object model used in RetSim the refund fraud scenario was implemented by the following setting: Estimate the average number of refunds per sale and the corresponding standard deviation. Use these statistics for simulating refunds in the RetSim model. Fraudulent salesmen will perform normal refunds, as well as fraudulent ones. The volume of fraudulent refunds can be modelled using a salesman specific parameter. The “red flag” for detection will in this case be a high number of refunds for a salesman. Similar to refund scenario, RetSim generates malicious coupon reduction/discounts and the analysis can also be performed in similar way as with refund fraud.