Insurance fraud costs South Africa (and the global insurance industry) billions of
Rands. Insurance claims fraud, which involves over-inflating claim amounts or fabricating
a loss to result in a claim settlement, makes up a substantial portion of this
cost. It would therefore be beneficial to the insurance industry to have a way of intelligently
identifying insurance claims fraud. Current strategies focus on identifying fraud
after the fact through methods such as auditing. These methods can be enhanced by
predicting whether claims are fraudulent before they get paid, instead of after payment
has already been made.
Techniques in the fields of data science and machine learning can be used to intelligently
predict insurance claims fraud, based on existing data. Because insurers have large sets
of data, it is suggested that Big Data be factored in when predicting insurance claims
fraud. However, new and proposed privacy legislation requires data scientists to be
mindful and consider privacy when mining users’ personal information.
The current research addresses the problems of insurance fraud, data bloat and information
privacy by proposing a framework, model and architecture. The proposed
framework contains the processes necessary to intelligently predict insurance claims
fraud. The model that is suggested can be used to predict insurance claims fraud. The
architecture shows software and hardware components that can be used to create a
prototype. The research as a whole discusses this prototype, how it was developed,
and how it was tested.