Intrusion Detection Model Evaluation Results Across Eight Scenarios

Evaluation results for machine learning models across eight distinct scenarios (S1–S8) assessing robustness to domain shift in network intrusion detection. The data originates from the TAN-IDS evaluation framework research by Dung Ha Thanh, published on figshare in April 2026. This 5.5 KB XLS file contains comparative performance metrics.

Use Cases

• Compare model F1-macro and recall scores across the eight in-dataset and cross-dataset evaluation scenarios (S1–S8).
• Analyze performance degradation metrics for attack-class detection under direct cross-dataset transfer conditions.
• Benchmark the effectiveness of domain-aware strategies like mixed-domain training and target-domain fine-tuning using the reported performance metrics.
• Evaluate the impact of a compact 8-dimensional NetFlow feature representation on model interoperability and reported accuracy.

Strengths

• Results derived from an evaluation framework tested with representative ML models and neural architectures, including a Transformer-based model.
• Formalizes key deployment conditions as eight explicit evaluation scenarios for structured comparison.
• License is CC-BY-4.0, permitting reuse and redistribution with attribution.

Limitations

• Dataset is extremely small at 5.5 KB, indicating it contains summary results, not raw traffic or feature data.
• Specific row count and detailed column schema (e.g., exact metric names) are not provided.
• The underlying network traffic datasets used to generate these results are not included.

Provenance

Source

Research output from the TAN-IDS evaluation framework study by Dung Ha Thanh, hosted on figshare.

Collection Method

Results generated from extensive experiments within the proposed TAN-IDS framework, unifying heterogeneous traffic datasets under an 8-feature NetFlow interface.

Time Range

null

Freshness

Last updated April 2026, indicating recent research.

Geography

null