Developing a BERT-Enhanced Blockchain Model for HealthInsurance Fraud Detection

Abstract

The health insurance fraud issue has been a significant challenge in the healthcaresystem. Worldwide organisations have been actively searching for effective strate-gies to either prevent or manage this problem. For this purpose, we are introducingHI-BERT (Health Insurance Bidirectional Encoder Representations from Transform-ers), which is based on BERT with adjustments made to vocabulary. We also useNaive Bayes as a supervised classification machine learning algorithm to label pre-scriptions by disease type. We check the similarity between the prescriptions andInternational Classification of Diseases (ICD) codes, which are designated by theWorld Health Organisation (WHO) to enhance the fraud detection process. Withthe objective of securing access to patients’ medical records, we have developed ahybrid blockchain in order to create a strong system for securely storing and retriev-ing data. Consequently, we can make sure that health information is protected to anextent that makes healthcare services more trustworthy. By integrating HI-BERTwith Naive Bayes, the performance is enhanced in detecting medical prescriptionsemantic similarity, and we get more accurate results compared to the basic BERT.Also, by using a hybrid blockchain instead of a single public blockchain, we increasedsecurity and reduced network load.