Lightweight Deep Learning-Based Model for Traffic Prediction in Fog-Enabled Dense Deployed IoT Networks

Ateya, Abdelhamied A.; Soliman, Naglaa F.; Alkanhel, Reem; Alhussan, Amel Ali; Muthanna, Ammar; Koucheryavy, Andrey

doi:10.1007/s42835-022-01314-w

Cited by 6 publications

(1 citation statement)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Generally, simple and efficient NNs are considered to be a data processing method with great potential when talking about network traffic prediction [12,13]. Particularly, such networks are highly suitable for deployment on edge devices, and their outstanding computational efficiency enables rapid processing of continuously changing traffic data, consistently providing real-time information for network traffic management and monitoring.…”

Section: Introductionmentioning

confidence: 99%

Fault-tolerant Echo State Networks for Mobile Network Traffic Prediction

Sun,

Gao,

Wang

et al. 2023

Preprint

View full text Add to dashboard Cite

Network traffic prediction is widely recognized as a potential solution for many functions of network security and management. Currently, numerous neural networks (NNs) are used in network traffic prediction tasks. When errors occur in the computational units implemented in NNs due to the effects of hardware failures, the results of arithmetic operations can be altered, and the prediction results may be affected by errors. This is unacceptable when NNs are used for network traffic prediction, because erroneous prediction results may lead to network congestion, uneven resource allocation, and other problems. Therefore, for NNs operating on mobile edge devices, an effective fault tolerance scheme needs to be embedded. A fault-tolerant model for fixed-type neuron failures in echo state networks (FtESN), is proposed in this paper, to improve its reliability in mobile network traffic prediction tasks. There are three specific contributions as follows: 1) We model the fixed-type faulty neurons and introduce the concept of degree for indexing. 2) A simple fixed fault tolerance mechanism is proposed for these faulty neurons. 3) The boundary conditions for reservoir neuron fault tolerance are proved by mathematical theoretical derivation. Experimental results show that the method is effective in recovering performance from fault patterns in terms of prediction accuracy, frequency-domain analysis, statistical distribution comparison, and short-term memory capacity (MC).

show abstract

Section: Introductionmentioning

confidence: 99%