Cyber Threat Detection in Federated Learning: A Secure, AI- Powered Approach Using KNN, GANs, and IOTA

Abstract

Cyber threats are dynamically changing and can easily break down traditional security architectures that rely on centralized models. Such models can be easily targeted by data breaches and computational overhead. This work introduces a novel, decentralized framework for the secure detection of cyber threats, with the integration of Federated Learning (FL), K- Nearest Neighbors (KNN), Generative Adversarial Networks (GANs), and IOTA Tangle, thereby improving anomaly detection while preserving privacy. It utilizes KNN for real-time anomaly detection, GAN for synthetic attack generation, and IOTA for secure communication in IoT environments. The experimental results show higher performance, with 97% accuracy, 95% precision, 96% recall, and 98% detection rate, bringing the false positive rate down to 3%. The system is also characterized by low latency, 35 ms, and high throughput, 250 transactions per second, surpassing traditional cybersecurity techniques. It demonstrates the real-time effectiveness of the AI-driven, blockchain-supported cybersecurity framework in detection and mitigation.