Healthcare fraud detection using adaptive learning and deep learning techniques
The healthcare industry faces huge losses due to the mismanagement of insurance transactions. Due to the development of public and private healthcare programs, many citizens receive better medical care benefits. Still, there is a need for financial transparency in these healthcare transactions, which has become a challenge. To ensure the delivery of more effective and higher-quality healthcare services, introducing healthcare fraudulent transactions prevention and detection tools in hospitals is necessary. In this paper, we propose how to inculcate a healthcare transaction monitoring system within an enterprise or organisation. Using machine and deep learning techniques, this research proposes a novel framework for analyzing health insurance data. Due to the complexity of medical information, detecting fraudulent transactions in the industry requires effort. Typically, patients, services, and providers (doctors, hospitals, pharmacies) are the main key elements of the healthcare ecosystem. As fraudsters continue to evolve their methods of conducting fraudulent transactions over time, an evolving fraud detection framework needs to be developed. Therefore, we proposed a framework that can identify fraud at the actor-level and further analyze the identified element (doctor, patient, and services) using an Anomaly transformer to evaluate the behavior of that particular identified element. Actor-level frauds are detected, 50% are at the patient level, 12% are at the service versus doctor level, 13% are at the service versus patient level, and 25% are at the physician level. Further, sequences of these elements are analyzed by the Anomaly transformer. All patient sequences’ anomaly scores are generated using a data-driven threshold, and fraudulent sequences are identified. Results of the Speciality-based Rule engine and the Anomaly transformers are compared to identify the anomaly finally. Once the frauds are identified, the proposed architecture enables the management to take disciplinary action against each involved element. The Accuracy of the proposed framework is 97%, The experimental results are validated using the insurance data of local hospital employees, and the domain expert has validated the detected fraud cases.
Item Type | Article |
---|---|
Additional information | © 2025 The Author(s). This is an open access article under the Creative Commons Attribution Non-Commercial No-Derivatives CC BY-NC-ND licence, https://creativecommons.org/licenses/by-nc-nd/4.0/ |
Keywords | sequence mining, insurance management, fraud detection, anomaly score, anomaly detection, anomaly rank, healthcare |
Date Deposited | 06 Jun 2025 08:59 |
Last Modified | 06 Jun 2025 08:59 |