An intelligent method of predicting insurance claims fraud

Abstract

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.