Federated Learning-Based Anomaly Detection Algorithm for Privacy-Preserving Health Information Systems

Abstract

Integrating privacy-preserving models with anomaly detection algorithms is an important component for secure patient information processing in health information systems. This paper describes a new anomaly detection algorithm based on Federated Learning to identify anomalous behaviors in health information systems, while ensuring privacy protection. Sensitive medical information stays decentralized; there is no raw data leaving local nodes, thus protecting patient privacy. The proposed algorithm improves the identification of security anomalies like unauthorized access to health information or disruptions of the operational status of health information systems based on privacy initiatives such as HIPAA. The study evaluates the feasibility of the proposed model in currently available real-world health datasets. The results confirm promising detection accuracy with low false positive rates and improved scalability between heterogeneous health institutions. The findings suggest that federated anomaly detection provides a credible option for privacy-preserving health data usage with adequate performance and safeguards against illness-related data leaks that exploit sensitive health data.