Blockchain-Assisted Federated Learning for Cybersecurity: Combining Isolation Forest, Variational Autoencoders, and Differential Privacy

Abstract

The complexity of the cyber threats dictates the need for strong, privacy-preserving mechanisms for anomaly detection. This paper introduces a new framework called BAFL, an integration of Isolation Forest and Variational Autoencoders combined with Differential
Privacy, for safe and scalable solutions in cybersecurity applications. Federated Learning allows distributed training across numerous clients without exposure of sensitive information, while the blockchain technology introduces trust and integrity in model updates. The addition of Differential Privacy (DP) raises security by being resistant to adversarial inferences; anomaly detection techniques Isolation Forest & VAE increase model accuracy in detecting threats. Performance evaluations indicate that the proposed model attains a 91.8% accuracy, 88.6% precision, and reduced the false positive rate to 5.3%, while outperforming traditional cybersecurity detection methods. This paper provides a scalable, privacy-preserving, and high- performance anomaly detection system which is appropriate for real-world cybersecurityapplications.