Hybrid Deep Learning for Traffic Classification in Low-Resource Networks

Gamsu, Michael and Chavula, Josiah (2025) Hybrid Deep Learning for Traffic Classification in Low-Resource Networks, Proceedings of 46th Annual conference of the South African Institute of Computer Scientists and Information Technologists (SAICSIT 2025), 17-18 July 2025, Durban, South Africa, Annual Conference South African Institute for Computer Scientists and Information Technologists, 406-422, SAICSIT.

Text online-Gamsu.pdf - Published Version Available under License Creative Commons Attribution. Download (847kB)

Abstract

Accurate network tra!c classification is essential for enhanc- ing Quality of Service (QoS) and optimizing tra!c engineering. How- ever, traditional classification techniques face significant challenges in e”ectively handling encrypted tra!c. Moreover, many machine learning approaches are either too computationally intensive or insu!ciently gen- eralized to be practical for community networks with limited resources. To address these constraints, this study investigates hybrid deep learn- ing architectures that combine Convolutional Neural Networks (CNNs) with Recurrent Neural Networks (RNNs)—specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. The evaluated models include 2D-CNN-LSTM, 2D-CNN-GRU, 1D-CNN-LSTM, and 1D-CNN-GRU, in addition to baseline models such as multilayer perceptrons (MLP), 2D-CNN, and 1D-CNN. Preliminary results indicate that hybrid models, particularly 1D-CNN-LSTM and 1D-CNN-GRU, achieve the highest classification precision. However, these models incur considerable computational overhead, limiting their applicability in resource-constrained environments. Conversely, 2D-CNN-based hybrids demonstrate a promising trade-off, offering competitive performance with lower computational demands, making them more suitable for deployment in community networks.

Item Type:	Conference paper
Uncontrolled Keywords:	Deep Learning Community Networks Traffic Classification Neural Networks
Subjects:	Networks > Network algorithms > Data path algorithms > Packet classification Networks > Network algorithms > Data path algorithms > Deep packet inspection Networks > Network performance evaluation > Network measurement
Date Deposited:	23 Jun 2026 12:05
Last Modified:	23 Jun 2026 12:05
URI:	https://pubs.cs.uct.ac.za/id/eprint/1786

Actions (login required)

View Item