GC-FIW NetFlow Research Data and Code
Description
Code, configurations, frozen aggregate metric summaries, and figure-reconstruction inputs supporting the manuscript 'Auditable Feature-Importance Allocation for Low-False-Positive NetFlow Anomaly Detection'. The package covers four public NetFlow v2 domains and includes five-seed exploratory summaries plus a prespecified confirmation on two domains. Raw traffic records, processed arrays, model checkpoints, and per-record anomaly scores are not redistributed. The original datasets remain available from their cited public DOI records. The deposit supports audit of reported aggregate values, implementation, configuration, and figure reconstruction; it is not an end-to-end binary replay of every training run.
Files
Steps to reproduce
1. Use Python 3.11 or 3.12. 2. Extract the ZIP archive. 3. Install the dependencies declared in pyproject.toml. 4. Run: PYTHONPATH=src python -m pytest -p no:cacheprovider -q tests/test_feature_importance.py tests/test_paper_assets.py 5. Validate frozen inputs with: python paper_preparation/generate_assets.py --check-only 6. Rebuild the English figures and tables with: python paper_preparation/generate_assets.py --language en --output-dir rebuilt_assets. These commands validate and redraw deposited aggregates; they do not retrain models or access source test records.
Institutions
- Shanghai Jiao Tong UniversityShanghai, Shanghai