A Novel Distributed Machine Learning Model to Detect Attacks on Edge Computing Network

To meet the growing number and variety of IoT devices in 5G and 6G network environments, the development of edge computing technology is a powerful strategy for offloading processes in data servers by processing at the network and nearby the user. Besides its benefits, several challenges related to decentralized operations for improving performance or security tasks have been identified. A new research direction for distributed operating solutions has emerged from these issues, leading to applying Distributed Machine Learning (DML) techniques for edge computing. It takes advantage of the capacity of edge devices to handle increased data volumes, reduce connection bottlenecks, and enhance data privacy. The designs of DML architectures have to use optimized algorithms (e.g., high accuracy and rapid convergence) and effectively use hardware resources to overcome large-scale problems. However, the trade-off between accuracy and data set volume is always the biggest challenge for practical scenarios. Hence, this paper proposes a novel attack detection model based on the DML technique to detect attacks at network edge devices. A modified voting algorithm is applied to core logic operation between sever and workers in a partition learning fashion. The results of numerical simulations on the UNSW-NB15 dataset have proved that our proposed model is suitable for edge computing and gives better attack detection results than other state of the art solutions.