Identity fraud has become a significant concern and has cost the Australian economy $2.65 billion in 2015-2016. Due to the possibility of significant financial gains, identity fraud has been steadily changing in nature from being primarily an activity performed by individuals into the domain of organised crime. This has turned it, according to Australian Crime Commission, into one of the six key "enablers" for organised crime.
One of its subsets, the synthetic identity fraud, has emerged only recently. The growth of synthetic identity fraud is closely related to business pressures financial organisations face and the increased use of online service delivery. A bank that can approve an online credit application without the need for human intervention will boost its bottom line significantly. But that approval process needs to be strong enough to compensate for the lack of human presence that normally allows for stronger controls such as analysing visual or verbal cues.
Synthetic identity fraud has been growing substantially and in 2017 the losses attributed to it were $16.8 billion, compared to $5 billion in 2014. This growth is likely to continue unabated due to inconsistencies across organisations in how identities are evaluated by the internal controls.
This study explores mechanisms aimed at increasing the detection of synthetic identities and minimising their impact on financial services sector and the economy. The study proposes a methodology for detecting synthetic identity fraud in credit applications. The methodology is based on analysing the behaviour of a credit applicant on social media, as well as analysing the answers provided by him or her to questions asked as part of the application process. The theoretical foundation for this research is based on the following theories: Human Identification Theory (IMT), Information Manipulation Theory (IMT), Interpersonal Deception Theory (IDT), Psychological Reactance Theory (PRT), and Fraud Triangle Theory (FTT). They underpin the study’s research question: How can Social Media Graph Analysis (SMGA) and Natural Language Processing (NLP) be used to improve detection of fraudulent credit applications based on synthetic identities? which has led to the development of a conceptual model and a prototype based on that model. The prototype implements routines to analyse the behaviour of an applicant on Facebook and Twitter and analyse the text answers related to his or her personal financial situation. The verification of the prototype was done through a series of tests with real and simulated data, while its validation was done by interviewing independent subject matter experts.
Additionally, the study has resulted in a number of contributions to academic literature and fraud detection practice. The academic contributions address the limitations of current synthetic identity fraud detection approaches. The practical contributions include a risk score calculator, using visualisation to assist integration with existing fraud detection processes, implementation of social media graph analysis, applying natural language processing in fraud detection, and summarising all results to be included into corporate processes.