A Novel Wavelet Transform and Deep Learning-Based Algorithm for Low-Latency Internet Traffic Classification

Accurate classification of low-latency Internet traffic is essential for real-time applications 1 such as video conferencing, online gaming, financial trading, and autonomous systems, where even millisecond-level delays can significantly degrade performance. Modern networks must dynamically allocate resources based on real-time traffic demands to ensure seamless Quality of Service (QoS). Streaming platforms like Netflix and YouTube already utilize adaptive bitrate algorithms to optimize playback based on network conditions, highlighting the need for intelligent traffic management. However, existing classification methods relying on raw temporal features or static statistical analyses struggle to capture the highly dynamic and bursty nature of low-latency traffic. This paper introduces a novel algorithmic framework that uniquely integrates wavelet transforms (WT) with artificial neural networks (ANNs) to address this gap. Unlike prior works, we systematically apply WT to commonly used temporal features such as throughput, slope, ratio, and moving averages transforming them into frequency-domain representations. This approach reveals hidden multi-scale patterns in low-latency traffic, akin to structured noise in signal processing, which traditional time-domain analyses often overlook. The transformed features are then used to train an ANN classifier, enabling precise distinction between low-latency and non-low-latency traffic. Our methodology diverges from conventional practices in two key ways: (1) Feature Enhancement: By applying WT to temporal features, we expose high-frequency components that correlate with rapid packet exchanges a hallmark of low-latency traffic. (2) Hybrid Architecture: The integration of WT with ANNs creates a dual-domain (time and frequency) analysis framework, enhancing classification robustness. Experiments demonstrate the algorithm’s superiority over existing methods, achieving 99.56% accuracy in distinguishing low-latency traffic (e.g., video conferencing) from FTP and video streaming. Even in complex scenarios with mixed traffic types, the model maintains 74.2–92.8% accuracy, outperforming benchmarks such as k-NN, CNNs, and LSTMs. For Internet Service Providers (ISPs), this approach offers a scalable solution to prioritize time-sensitive traffic and improve real-time network performance without relying on deep packet inspection. By bridging signal processing and deep learning, our work advances the state-of-the-art in traffic classification, ensuring efficient bandwidth utilization and enhanced QoS in increasingly heterogeneous network environments.