Nslkdd Fetfids Eval

nslkdd-fetfids-eval

FetFIDS: A Feature Embedding Attention based Federated Network Intrusion Detection Algorithm — Ghosh et al. (2025) (arXiv:2508.09056, 2025)

What this evaluates

Evaluates a federated transformer-based intrusion detection model on network traffic data to classify benign and malicious packets across five attack categories under a realistic class-imbalanced, distributed setting.

Datasets

• NSLKDD — total ?; splits: train (-1), val (-1), test (-1)

Metrics

• detection performance (primary) — range: percent
  • Not explicitly defined in the provided text; typically refers to classification accuracy or F1-score in intrusion detection contexts.

Input / output format

Input: 41-feature network traffic packet records

Output: Classification into one of 5 classes: Benign, DoS, U2R, R2L, or Probing

Scoring recipe

# Not specified in provided text. Standard classification accuracy is assumed:
# def score(preds, gold):
#     return sum(p == g for p, g in zip(preds, gold)) / len(gold)

Common pitfalls

• Class imbalance is intentionally preserved across federated devices to emulate realistic edge scenarios, so data augmentation is not applied.
• The dataset is split into fixed percentages (85/7.5/7.5) rather than using the original NSL-KDD fixed train/test splits.

Evidence (verbatim from paper)

achieving superior detection performance in edge-based federated learning settings by focusing on attack-specific feature patterns rather than positional context.

Citation

@misc{ghosh2025fetfids,
  title={FetFIDS: A Feature Embedding Attention based Federated Network Intrusion Detection Algorithm},
  author={Ghosh et al. (2025)},
  year={2025},
  note={arXiv:2508.09056}
}

• arXiv: 2508.09056