Abstract
The dramatic increase of IOT devices and decentralized networks has aggravated the potential entry points of cyber threats, resulting in a demand for efficient Intrusion Detection Systems (IDS). Centrally controlled traditional models of learning are affected by issues of latency, limits of the bandwidth, and privacy leakage. To address these problems, the present work put forward a Local Computing Enhanced Federated Learning (EC-FL) framework combined with Autoencoder-based Machine Learning models to achieve on-the-fly intrusion detection. The proposed schema spreads out model training across the edge devices, thus allowing collaborative learning without the need of sharing raw data. Comparing cloud-based IDS, the experiments conducted on standard datasets like NSL-KDD and CICIDS2017 depict a better detection rate (96.7%), lower training time (31%), and lesser communication overhead. The method proposed is instrumental in augmenting scalability, ensuring data privacy, and enhancing the real-time anomaly detection in the interconnected world of IoT network infrastructures.

This publication has 11 references indexed in Scilit: