Work

AI Driven Framework for Need-Based Insurance Plans Generation and Anomaly Detection Using Deep Learning Techniques

Year: 2025

Type: article

Abstract: Traditional health insurance models lack flexibility in insurance coverage, as these models suggest fixed insurance policies and plans for individuals and enterprises. As a result, the insured employee is not adequately covered and has to bear additional expenses from his own pocket over and above the claim amount. Enterprises have to pay same premium amount for all employees even for those who utilize healthcare insurance services infrequently. This results into inadequate cost distribution. The manual and static evaluations and verification methods are used for processing insurance claims which may not fully mitigate the risks of entertaining fraudulent claims and ultimately enterprises and insurance companies may bear additional financial burdens. A transformative approach is needed, which can integrate insurance packages and anomaly detection techniques for prevention and timely detection of fraud, risk evaluation and claim optimization. The proposed framework introduces an efficient way to predict health insurance premiums and generate customizable insurance plans using Recurrent Neural Networks (RNNs) and clustering techniques. The proposed methodology attempts to predict the insurance premium amount and splits patients into three categories of risk- Low, Moderate, and High Risk-depending upon how they utilize healthcare services. These categorizations help to create dynamic and need-based insurance policies instead of the rigid designation-based policies. Moreover, to prevent and to monitor fraudulent practices in healthcare, we have applied LSTM-based Anomaly Transformer and Generative adversarial networks (GANs) and compared the results from both models. Experimental results show that Anomaly transformer generates better results as compared to GANs. The proposed framework efficiently detects anomalies in employee and physician behavior, with a 0.0917 validation loss and validation accuracy of 96.4%. The research demonstrates that the proposed AI driven framework can perform both coverage optimization using multiple patient data sets, as well as decision support in healthcare systems.

Source: IEEE Access

DOI: 10.1109/access.2025.3583562