Optimized Federated Learning for Cybersecurity: Integrating Split Learning, Graph Neural Networks, and Hashgraph Technology

Abstract

The complexity of the cybersecurity threats today requires more effective defense mechanisms 
which can detect malware in real time and protect the data. The paper presents a federated learning 
framework optimized through split learning that integrates graph neural networks and hashgraph 
technology in order to harden cybersecurity systems. The proposed model achieves 98% threat 
detection accuracy with a reduction of false positives to 2%, swift detection latency of 30 
milliseconds, and impressive throughput of 250 transactions per second. With the help of GNNs 
for robust anomaly detection, the system can effectively identify complex cyber threats. Moreover, 
Hashgraph Technology provides secure and transparent data exchange that increases scalability 
and efficiency. These integrated technologies address challenges in distributed and decentralized 
environments, which makes the framework highly adaptable to IoT, cloud, and edge computing. 
The experimental evaluations demonstrated the superiority of the framework over conventional 
models by underlining the capability for real-time threat mitigation and enhanced privacy 
preservation, which makes it suitable for modern applications in cybersecurity.