Mixed-Domain Training Results for NetFlow-Based Intrusion Detection

Results from experiments (S17–S24) evaluating machine learning models for intrusion detection under domain shift. The 5.5 KB dataset, created by Dung Ha Thanh, was last updated in April 2026. It contains performance metrics from the TAN IDS evaluation framework.

Use Cases

• Analyze cross-dataset performance degradation using the 8-dimensional NetFlow feature interface.
• Compare model robustness between in-dataset evaluation and direct cross-dataset transfer scenarios.
• Evaluate the effectiveness of mixed-domain training strategies on attack-class recall and F1-macro scores.
• Assess lightweight target-domain fine-tuning performance with metrics from experiments using only 5% labeled data.

Strengths

• Results derived from a formalized evaluation framework (TAN-IDS) designed for reproducible, deployment-centric analysis.
• Includes performance metrics for key scenarios like mixed-domain training and fine-tuning, with reported F1-macro scores exceeding 95% in some cases.

Limitations

• Extremely small dataset at 5.5 KB, indicating limited scope and likely containing only summary results, not raw data.
• Unknown row count and specific column structure, limiting detailed statistical analysis.

Provenance

Source

figshare, author Dung Ha Thanh.

Collection Method

Generated from experiments within the TAN-IDS evaluation framework using representative ML models and neural architectures.

Time Range

null

Freshness

Last updated in April 2026.

Geography

null