Adversarial Robustness Evaluation for Cybersecurity Datasets

A 5.5 KB Excel dataset by Sudip Saha, last updated in 2026, containing results from adversarial robustness experiments. The underlying research reports up to 97.88% accuracy on clean data and maintains 84.9% accuracy under FGSM attacks. It compares a hybrid adversarially-trained model against CNN and LSTM baselines on reconnaissance, shellcode, and worms datasets.

Use Cases

• Benchmarking adversarial defense methods based on reported accuracy metrics for FGSM and PGD attacks.
• Analyzing the trade-off between clean accuracy and adversarial robustness based on the described loss components.
• Studying model convergence stability and calibration error based on the reported training curve metrics.

Strengths

• Includes specific performance metrics: 97.88% clean accuracy, 84.9% FGSM robustness, and 81.75% PGD robustness.
• Compares model performance against established CNN and LSTM baselines, showing a 6–10 percentage point improvement.
• Reports secondary metrics like a 30% reduction in expected calibration error and improvements in gradient norm decay.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small at 5.5 KB, indicating limited scope, likely containing summary results rather than raw data.

Provenance

Source

figshare, author Sudip Saha

Collection Method

Likely contains experimental results from a research paper on a hybrid adversarial training framework.

Freshness

Last updated 2026-06-01 17:55:14; freshness should be verified.