Federated Learning for Cybersecurity: Decentralized Threat Detection in Large Networks

Abstract

Cybersecurity has become one of the most critical challenges in modern computing as cyber
threats increase in sophistication and frequency. Traditional centralized security models suffer
from several weaknesses, including privacy concerns, high communication costs, and
susceptibility to attacks targeting a single point of failure. Federated Learning (FL) presents a
novel approach by enabling distributed training of threat detection models across multiple devices
without transferring raw data, ensuring privacy and efficiency. This paper explores the
application of Federated Learning for cybersecurity, specifically in decentralized threat detection
across large-scale networks. The study evaluates the efficiency of FL models in identifying various
cyber threats, including malware, phishing attempts, denial-of-service (DoS) attacks, and
advanced persistent threats (APTs). The proposed FL-based cybersecurity framework is
compared with traditional centralized security models and conventional intrusion detection
systems (IDS), highlighting its higher detection accuracy, lower false positive rates, and improved
privacy protection. Experimental results indicate that FL-based threat detection reduces the risk
of data breaches, increases model adaptability in dynamic environments, and provides a scalable
approach to securing enterprise and IoT networks.