Insurance Fraud Detection using Graphical Neural Networks (GNNs) and Machine Learning

Insurance fraud schemes are becoming more and more sophisticated as their perpetrators gain knowledge of the fraud prevention methodologies used by insurers. It becomes increasingly difficult for existing state-of-the-art fraud detection applications to persist in using the relational data models that underpin most conventional technologies because fraudsters are well aware of how best to make claims that avoid detection.

To combat this in our novel development, it is necessary to store insurance claim data in such a way that the connections between entities in different claims (such as people, vehicles, addresses or telephone numbers) are represented as comprehensively as possible.

The problem that this technology will solve is how to leverage the unique benefits of a graph database for insurance fraud detection in real-time. Graphical Neural Networks (GNNs) are a new development in the world of machine learning and their application in the area of insurance fraud is in its infancy. Our solution implementing GNNs include, but not limited to, identifying syndicates through community detection, finding hidden links to entities of interest, and usual both contextual and data to identify staged accidents and thefts.

No competitors in the insurance fraud area are using graph networks and GNNs in this way to find hidden connections between entities, identifying communities, detecting anomalies, and classifying individual entities.

KBBS novel technology operates as a stand-alone product or as an integrated part of a customer solution, the overarching process comprising ingestion of claim data into the graph database and the execution of fraud detection models and rules. Real-time processing enables incorporation into customer workflows and applications with individual claims being submitted as requests to an API such as Guidewire Claims system.