Which Deep Learner? A Systematic Evaluation of Advanced Deep Forecasting Models Accuracy and Efficiency for Network Traffic Prediction

TL;DR

This work tackles the challenge of reliable network traffic prediction by conducting a large-scale, efficiency-aware benchmark of 12 deep forecasting models across four real datasets and multiple time scales. It systematically analyzes generalization, data efficiency, and deployment-resource efficiency while accounting for anomalies, missing data, and external factors. The study finds that no single model dominates all scenarios, but highlights data-efficient architectures like DLinear and PatchTST, whose performance remains strong across scales with favorable resource profiles. The results provide actionable guidance for deploying NTP systems and outline directions for future research toward efficient, robust, and scalable forecasting in network environments.

Abstract

Network traffic prediction is essential for automating modern network management. It is a difficult time series forecasting (TSF) problem that has been addressed by Deep Learning (DL) models due to their ability to capture complex patterns. Advances in forecasting, from sophisticated transformer architectures to simple linear models, have improved performance across diverse prediction tasks. However, given the variability of network traffic across network environments and traffic series timescales, it is essential to identify effective deployment choices and modeling directions for network traffic prediction. This study systematically identify and evaluates twelve advanced TSF models -- including transformer-based and traditional DL approaches, each with unique advantages for network traffic prediction -- against three statistical baselines on four real traffic datasets, across multiple time scales and horizons, assessing performance, robustness to anomalies, data gaps, external factors, data efficiency, and resource efficiency in terms of time, memory, and energy. Results highlight performance regimes, efficiency thresholds, and promising architectures that balance accuracy and efficiency, demonstrating robustness to traffic challenges and suggesting new directions beyond traditional RNNs.

Figures (13)

• Figure 1: Characteristics of network traffic time series across multiple time scales. (a) Multi-Periodicity at the hourly scale. (b) Fluctuations at the 10-minute scale. (c) Anomalies at the 5-minute scale. (d) Missing data on the 15-minute scale. (e) Variations due to External factors (windspeedwindspeed) at the hourly scale.
• Figure 2: Our proposed evaluation framework. The upper part shows core components, and the bottom strip illustrates the unified training and evaluation process.
• Figure 3: CWT Spectrograms and Corresponding FFT Plots for the $4$ Traffic Datasets. Each dataset is represented by two figures: one for the CWT spectrogram and one for the corresponding FFT plot. The bottom row presents the FFT plots, with marked peaks indicating the dominant daily and weekly frequencies. The top row shows the corresponding CWT spectrograms, highlighting the time-frequency distribution of these frequencies.
• Figure 4: Missingness maps for three dataset pairs illustrate two missing-at-random patterns with different gap durations. Each pair has an upper map showing short-gap missingness (missing intervals up to one day), and a lower map showing long-gap missingness (intervals over several days) week.
• Figure 5: Quantitative evaluation of model performance across 11 datasets. Models on the x-axis are arranged in chronological order, and for each model, 11 bars report the NRMSE for each dataset, with error bars indicating the standard deviation across repeated runs. This comparison highlights differences in accuracy among models as well as the degree of performance variability across datasets.