1.5 KiB
1.5 KiB
title, date, draft, tags, categories, summary, author
| title | date | draft | tags | categories | summary | author | ||||
|---|---|---|---|---|---|---|---|---|---|---|
| [bojemoi] feat(ml-threat): richer features + per-class F1/recall metrics | 2026-07-08T00:28:08+02:00 | false |
|
|
Commit 1e80d42 par grafana-watcher dans bojemoi | grafana-watcher |
Commit 1e80d42
| Repository | bojemoi |
| Branch | main |
| Author | grafana-watcher |
| Hash | 1e80d42b8caf869f8426277f40a75e50ba18ebea |
Description
Features (#2):
- SQL: add web_port_count, db_port_count, remote_access_count (CASE WHEN port IN (...))
- Replace 4 dead features (report_count/spam/ssl/dns) with web/db/remote/port_entropy
- port_entropy: Shannon entropy over 4 service categories (0-2 bits)
- Updated reputation heuristic: weighted by service sensitivity (db5, remote4, web*3)
- Synthetic data updated to match new 10-feature layout
Metrics (#1):
- Log precision/recall/F1 per class after each retrain
- Return per_class dict in /models/retrain API response
- Result: malicious recall=0.828 (17% miss rate, now measurable and actionable)
Co-Authored-By: Claude Sonnet 4.6 noreply@anthropic.com
Files Changed
M ml-threat/api.py
M ml-threat/train.py
M ml-threat/train_from_msf.py
Diff Summary
ml-threat/api.py | 1 +
ml-threat/train.py | 98 ++++++++++++++++-----------------
ml-threat/train_from_msf.py | 131 +++++++++++++++++++++++++++++++++++---------
3 files changed, 152 insertions(+), 78 deletions(-)