Ablation Study of a Hybrid Adversarial Defense Model for Cybersecurity Datasets

Sudip Saha's ablation study evaluates a hybrid adversarially-trained deep learning framework. The dataset likely contains experimental results comparing model performance on clean and adversarial data, including metrics like accuracy, calibration error, and gradient dynamics. It was last updated on June 1, 2026.

Use Cases

• Benchmarking adversarial robustness based on reported accuracy under FGSM and PGD attacks
• Analyzing the trade-off between clean accuracy and adversarial robustness based on the described hybrid framework
• Studying model convergence stability based on mentioned training curves and gradient norm decay
• Comparing defense mechanisms against baseline CNN and LSTM models based on the described performance gap

Strengths

• Includes specific performance metrics: up to 97.88% clean accuracy and 84.9% accuracy under FGSM attacks
• Reports a 30% reduction in expected calibration error, indicating improved model reliability
• Compares results against established CNN and LSTM baselines, showing a 6–10 percentage point improvement in robustness

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• The 5.5 KB file size suggests a very limited scope, likely containing summary results rather than raw experimental data

Provenance

Source

Sudip Saha via figshare

Collection Method

Likely generated from experiments on reconnaissance, shellcode, and worms datasets as part of an academic study.

Freshness

Last updated 2026-06-01 17:55:15; freshness should be verified