Training Time and Efficiency Trade-offs for Network Intrusion Detection Models

Shailendra Mishra's framework evaluates Intrusion Detection Systems (IDS) using harmonized features from the NSL-KDD and CICIDS2017 datasets. The work benchmarks supervised, unsupervised, deep learning, and ensemble models, reporting a Random Forest model achieving 98.0% accuracy and 97.0% F1-score on the harmonized data. The dataset, last updated in April 2026, is a 5.5 KB Excel file detailing experimental results and trade-offs.

Use Cases

• Benchmarking machine learning models for intrusion detection based on harmonized NSL-KDD and CICIDS2017 features.
• Analyzing the impact of feature harmonization on model performance using ablation analysis.
• Comparing training time and efficiency trade-offs across supervised, unsupervised, deep learning, and ensemble models.
• Evaluating statistical validation methods for IDS, such as Wilcoxon signed-rank and McNemar's tests.

Strengths

• Provides a unified framework with explicit performance metrics, including 98.0% accuracy and 97.0% F1-score for a Random Forest model.
• Includes a cryptographic logging mechanism using SHA-256 hash chaining for tamper-evident traceability.
• Benchmarks multiple model types (supervised, unsupervised, deep learning, ensemble) on two established network intrusion datasets.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The feature harmonization approach is based on manual alignment and may not be effective for highly heterogeneous datasets.

Provenance

Source

figshare, author Shailendra Mishra

Collection Method

Experimental results from a proposed evaluation framework for IDS.

Time Range

Experimental timeframe not specified; dataset updated April 2026.

Freshness

Last updated 2026-04-20 17:36:43; freshness should be verified.

Geography

Geographic coverage not specified.