Đã lưu trong:
| Tác giả chính: | |
|---|---|
| Định dạng: | Recurso digital |
| Ngôn ngữ: | Tiếng Anh |
| Được phát hành: |
Zenodo
2026
|
| Những chủ đề: | |
| Truy cập trực tuyến: | https://doi.org/10.5281/zenodo.19452335 |
| Các nhãn: |
Thêm thẻ
Không có thẻ, Là người đầu tiên thẻ bản ghi này!
|
Mục lục:
- <p>Healthcare fraud detection is a crucial research area due to its significant impact on rising medical costs and the overall integrity of healthcare systems. This paper presents an enhanced and deployable fraud detection framework that integrates multiple predictive models into a unified platform. A Flask-based web application was implemented to load and execute several models, including Convolutional Neural Network (CNN), Transformer, Graph Neural Network (GNN), Autoencoder, Random Forest, Decision Tree, XGBoost, and a hybrid model that combines CNN, Transformer, and GNN feature representations with an Autoencoder anomaly score and XGBoost for final classification. GNN captures network-level fraud patterns by modeling relationships between providers, patients, and claims as a graph, enabling detection of fraud rings and collusion networks. The Autoencoder functions as an unsupervised anomaly detector, learning the distribution of legitimate claims and flagging high reconstruction error as potential fraud. The system preprocesses incoming healthcare claims data, encodes categorical variables, and aligns features with the training dataset before generating predictions. To ensure transparency and trust, SHAP explainability is integrated, enabling visualization of the most influential features contributing to fraud predictions. Experimental results show that the hybrid model consistently outperforms individual models by capturing local, global, networklevel, and anomalous feature patterns simultaneously. The system offers an accurate, interpretable, scalable, and practical solution for real-world healthcare fraud detection</p>