Secure and explainable fraud detection in healthcare claims using blockchain-based machine learning

Area of Science:

• Health Informatics
• Blockchain Technology
• Machine Learning

Background:

• Healthcare insurance fraud incurs significant financial losses and erodes trust.
• Traditional fraud detection methods lack transparency and adaptability due to centralized systems and opaque models.
• Existing approaches struggle with data integrity and effective auditing of claims.

Purpose of the Study:

• To develop a secure and explainable framework for healthcare fraud detection.
• To integrate blockchain technology for tamper-resistant data storage and auditing.
• To enhance machine learning model transparency and adaptability in identifying fraudulent activities.

Main Methods:

• A permissioned blockchain was utilized for decentralized, immutable storage of claim evidence.
• A stacking ensemble model, incorporating LightGBM and XGBoost, was employed for fraud detection.
• SHAP (SHapley Additive exPlanations) was used for interpreting model predictions and providing feature-level explanations.
• Blockchain securely anchored hashed prediction and explanation records for verifiable auditing.

Main Results:

• The Voting classifier achieved a high Average Precision (AP) of 0.746, demonstrating effective fraud identification.
• The proposed stacking framework showed strong performance with 0.940 accuracy, 0.951 AUC, and 0.733 AP.
• SHAP values provided stable, feature-level explanations, enhancing model interpretability.
• The blockchain layer ensured secure, tamper-resistant auditing with minimal latency.

Conclusions:

• Integrating explainable machine learning with blockchain offers a trustworthy solution for healthcare fraud detection.
• The proposed framework enhances transparency, adaptability, and data integrity in combating insurance fraud.
• This approach supports auditable and verifiable identification of suspicious provider behavior in healthcare claims.