Graph Neural Network Models for Detecting Fraudulent Insurance Claims in Healthcare Systems

Fraudulent insurance claims in healthcare systems represent a persistent and costly challenge, undermining the efficiency, equity, and sustainability of healthcare delivery worldwide. Traditional approaches to fraud detection, including rule-based systems and statistical models, have provided valuable early insights but often fail to capture the complex, relational, and evolving nature of fraudulent behavior. This study addresses these limitations by investigating the application of Graph Neural Networks (GNNs) as an advanced analytical framework for detecting fraudulent claims. By representing patients, providers, and claims as interconnected nodes within graph structures, GNNs leverage relational dependencies and structural patterns that are frequently overlooked by conventional models. The research employed a mixed-methods design, combining quantitative experimentation with multiple GNN architectures—such as Graph Convolutional Networks, Graph Attention Networks, and GraphSAGE—with qualitative insights gathered from healthcare fraud investigators to evaluate interpretability and usability. The quantitative findings revealed that GNN models consistently achieved higher accuracy, precision, and recall compared to traditional classifiers, while the qualitative analysis highlighted the importance of interpretability and visualization in building stakeholder trust and improving investigative efficiency. Importantly, the literature review synthesized evidence from 112 peer-reviewed studies, providing a comprehensive overview of existing fraud detection methods and situating GNNs within the broader progression of healthcare fraud research. The results underscore that GNNs not only advance the technical accuracy of fraud detection but also offer practical tools for detecting collusive fraud networks and managing large-scale, heterogeneous claims data. This study contributes to both academic discourse and practical applications by demonstrating that GNNs are uniquely positioned to enhance fraud detection in healthcare insurance systems through their capacity to integrate relational learning, scalability, and operational transparency.