Designing artificial intelligence -based fraud detection systems capable of handling dynamic threat environments

The phenomenal growth of online services with Internet connectivity has led to major changes in the way people conduct business, increasing their frequency and lowering operating costs. However, this reliance on online services has also led to increased incidences of fraud originating from Internet connectivity and by centralizing agencies responsible for revenue collection. This has led to significant amounts of losses on individuals and industries. Fraud occurs because of the anonymity enabled by the Internet, integration of online services, and the movement of funds from one place to another without adequate identification. Emerging technologies such as mobile devices with high computing power and availability of vast amounts of personal information on the Internet make it easier for people to commit fraud. Mobile phone fraud, which mainly deals with fake account creation and abuse of free trial services and promotions, incurs huge losses on mobile carriers. Identity theft by online criminals may result in significant amounts of a person’s identity information being sold online (Phua et al., 2010; Bahnsen et al., 2016; Fiore et al., 2019). Therefore, there is a real need for efficient automated online fraud detection to prevent such misuse and losses.

Fraud detection is the latest challenge to the burgeoning area of processes that detect invalid or anomalous activities occurring in systems or organizations, referred to as anomaly detection. Detecting a fraudulent transaction or activity is a problem of identification of humans who are infeasibly distinct from each other, not a normal activity like money laundering through banking systems. These unusual activities are expected to occur in small amounts of the total activity, and yet some of these unusual activities can lead to huge disasters if they are missed, hence automatic online fraud detection is thus a very challenging problem (West & Bhattacharya, 2016; Roy et al., 2018).