Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data Augmentation for Traffic Classification

Chao Wang, Alessandro Finamore, Pietro Michiardi, Massimo Gallo, Dario Rossi

TL;DR

This paper tackles the underexplored use of data augmentation in traffic classification by systematically benchmarking 18 hand-crafted augmentations across three packet-time-series TC datasets. It reveals that sequence-order and masking augmentations outperform amplitude-based ones and that analyzing latent-space geometry helps explain when augmentations improve or harm performance. The study also investigates how augmentation strategy (batching), dataset characteristics, and combining augmentations influence results, finding that there is no single superior method and gains are dataset-dependent. The authors argue for a potential shift toward generative-model-based augmentation to efficiently navigate the design space and improve TC generalization, while also outlining practical limitations and future directions. Overall, the work provides a comprehensive framework for evaluating DA in TC and highlights concrete, actionable insights for improving TC systems through data-centric approaches.

Abstract

Data Augmentation (DA) -- enriching training data by adding synthetic samples -- is a technique widely adopted in Computer Vision (CV) and Natural Language Processing (NLP) tasks to improve models performance. Yet, DA has struggled to gain traction in networking contexts, particularly in Traffic Classification (TC) tasks. In this work, we fulfill this gap by benchmarking 18 augmentation functions applied to 3 TC datasets using packet time series as input representation and considering a variety of training conditions. Our results show that (i) DA can reap benefits previously unexplored, (ii) augmentations acting on time series sequence order and masking are better suited for TC than amplitude augmentations and (iii) basic models latent space analysis can help understanding the positive/negative effects of augmentations on classification performance.

Data Augmentation for Traffic Classification

TL;DR

This paper tackles the underexplored use of data augmentation in traffic classification by systematically benchmarking 18 hand-crafted augmentations across three packet-time-series TC datasets. It reveals that sequence-order and masking augmentations outperform amplitude-based ones and that analyzing latent-space geometry helps explain when augmentations improve or harm performance. The study also investigates how augmentation strategy (batching), dataset characteristics, and combining augmentations influence results, finding that there is no single superior method and gains are dataset-dependent. The authors argue for a potential shift toward generative-model-based augmentation to efficiently navigate the design space and improve TC generalization, while also outlining practical limitations and future directions. Overall, the work provides a comprehensive framework for evaluating DA in TC and highlights concrete, actionable insights for improving TC systems through data-centric approaches.

Abstract

Data Augmentation (DA) -- enriching training data by adding synthetic samples -- is a technique widely adopted in Computer Vision (CV) and Natural Language Processing (NLP) tasks to improve models performance. Yet, DA has struggled to gain traction in networking contexts, particularly in Traffic Classification (TC) tasks. In this work, we fulfill this gap by benchmarking 18 augmentation functions applied to 3 TC datasets using packet time series as input representation and considering a variety of training conditions. Our results show that (i) DA can reap benefits previously unexplored, (ii) augmentations acting on time series sequence order and masking are better suited for TC than amplitude augmentations and (iii) basic models latent space analysis can help understanding the positive/negative effects of augmentations on classification performance.
Paper Structure (30 sections, 8 figures, 6 tables)

This paper contains 30 sections, 8 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background and related work
  3. Data augmentation in traditional machine learning tasks
  4. Input space transformations.
  5. Latent space transformations.
  6. Generative models.
  7. Data augmentation in traffic classification
  8. Hand-crafted DA.
  9. Generative models.
  10. Design space
  11. Search space.
  12. Quantifying good variety.
  13. Training loss.
  14. Linking generative models to classifiers.
  15. Our goals and methodology
  16. ...and 15 more sections

Figures (8)

  • Figure 1: Input sample ${\bf x}$ shape and related notation.
  • Figure 2: Training batch creation policies.
  • Figure 3: Augmentations rank and critical distance (G1).
  • Figure 4: Comparing Replace, Inject and Pre-augment batch creation policies (G2).
  • Figure 5: Investigating train, augmented and test samples relationships (G3).
  • ...and 3 more figures