Generative AI with Big Data for Better Detection of Fraud in Medical Claims

Generative AI refers to a type of algorithms that can generate new content. This can be text, images, or any other form of data. While Large Language Models (LLMs) are a specific type o f generative A I that focuses o n 1 anguage, they are basically trained on massive textual input to understand and generate human-like text. This paper addresses the critical challenge of fraud detection in medical insurance claims, a pervasive issue causing significant financiallo sses in healthcare. This work is focused on devising a robust, automated system for detecting fraudulent activities. Where an integration of Generative AI, specifically L LM i s implemented with B ig Data processing frameworks to enhancements in fraud detection in medical claims. Each LLM was used as an embedding layer that transforms textual features of a real-world insurance claim data into numerical representations. These claims data has been collected from countries belonging to the Mena region. The results show advantages towards LLMs that were trained on specialized medical contexts as they show better capability of understanding medical expressions which reflects model's performance. Applying further sampling techniques such as class weight and up-sampling did not have a significant impact on the LLMs performance, with a little better performance for class weight. Gemini showed advantages over BERT medical language models on most experiments by achieving 90.44% of classification accuracy.