Federated learning-based IoT intrusion detection on non-IID data

Huang, Wenxuan, Tiropanis, Thanassis and Konstantinidis, Georgios (2022) Federated learning-based IoT intrusion detection on non-IID data. In, Internet of Things. GIoTS 2022. (Lecture Notes in Computer Science, 13533) Springer Cham, 326–337. (In Press) (doi:10.1007/978-3-031-20936-9_26).

Record type: Book Section

Abstract

Federated learning allows multiple parties to jointly train and update machine learning models without shared data to a central server, which is particularly suitable for intrusion detection in IoT environments. However, data on each node in the IoT scenario are usually not independently and identically distributed (IID), which poses an additional challenge to the convergence and speed of federated learning. To address this problem, there have been proposals to have nodes share a portion of their data centrally, which can still raise privacy concerns. In this paper we propose a strategy to improve the training of non-IID data by allowing mutually trusted nodes to self-aggregate into clusters of trust that will participate in federated learning as peers. We experiment with an up-to-date IoT dataset, Aposemat IoT-23 (IoT-23 for short) and show that using this strategy is considerably more accurate federated learning, comparably accurate to proposals that envisage central sharing a portion of node data, and comparable to centralised machine learning accuracy

Text

Federated Learning based IoT intrusion detection on non IID data - Accepted Manuscript

Restricted to Repository staff only

Request a copy