alirezaaminzadeh's picture
Publish LightGBM triage model with isotonic calibrator
1cddaaf verified
|
Raw
History Blame Contribute Delete
1.17 kB
metadata
license: mit
tags:
  - malware
  - lightgbm
  - static-analysis
  - ember
  - cybersecurity
library_name: lightgbm

EMBER Malware Triage — LightGBM Classifier

Binary classifier for static PE malware triage using EMBER-inspired 256-dimensional feature vectors.

Intended Use

  • Primary: SOC Tier-1 static triage of unknown PE files
  • Secondary: Research and benchmarking on EMBER-style features
  • Out of scope: Malware execution, sandbox detonation, exploit generation

Model Details

Property Value
Architecture LightGBM gradient boosted trees
Features 256-dim EMBER-inspired vector
Output Malware probability (0–1)
Calibration Isotonic regression on holdout
Training data Synthetic reproducible benchmark

Limitations

  • Trained on synthetic features — fine-tune for production
  • Static analysis cannot detect runtime-only behavior
  • Not a substitute for sandbox analysis on high-risk samples

Bias & Safety

  • No real malware samples in training bundle
  • Defensive use only
  • Human analyst review required for enforcement actions