In the past few decades, machine learning has revolutionized data processing for large-scale applications. Simultaneously, increasing privacy threats in trending applications led to the redesign of classical data training models. In particular, classical machine learning involves centralized data training, where the data is gathered, and the entire training process executes at the central server. Industry 4.0 allows the appearance of Internet of Things-based transactive energy system (IoTES) that involves new services with a number of independent distributed systems. These systems produce bulk data that is heterogeneous and they are prone to cyber-attacks, especially stealthy false data injection attacks (FDIAs). Lossy networks (RPL) security, intrusion detection (ID) is crucial in this area, considering that it is highly vulnerable to attacks, especially those executed by an insider. Although a lot of literature suggests the use of ID systems (IDSs) by applying a variety of techniques, there is relatively little literature offering insight into where the IDSs fall within the RPL topology. The gap in this study will be bridged by aggressively comparing three ID architectures in terms of central and distributed location and on several dimensions, including effectiveness, cost, privacy, and security. The results are supported by the overwhelming contribution of attacker position and IDS-to-attacker distance towards the detection. Therefore, in addition to ascertaining the effectiveness of the old ID systems, the research also probes how federated learning (FL) can enhance ID in the RPL networks. The aspect of the decentralized model training approach in FL can overcome the effect of attacker-position on the performance of an IDS system by making sure that information that is considered to be pertinent in the context of an attack is gathered at the node along with the IDS system, irrespective of its proximity to the potential attackers. In addition, the approach not only eliminates security issues, but it also reduces communication overhead between the ID nodes. This will mean that FL will lower the rate of large-scale data transfer and thereby eliminate the consequences of packet loss and latency that any lossy network will cause. Also, the gap filled by the research is the impact of local data sharing on FL performance and how it is possible to balance the effectiveness with security. The proposed computing method can be computed in parallel and allows detecting the stealthy FDIA on all the nodes without any failure. The simulation experiments support the suggestion that the scheme under consideration is superior to the state-of-the-art approaches in terms of detection accuracy and the complexity of computation when using a distributed environment and ensuring the data privacy of the messages. Keywords: Quantum Computing, Federated Learning, Machine Learning, Learning Process, Machine Learning Models, Internet Of Things, Transfer Learning
Ayub et al. (Sat,) studied this question.