metadata

license: apache-2.0
datasets:
  - asobirov/dga-preprocessed
  - harpomaxx/dga-detection
  - YangYang-Research/dga-detection
metrics:
  - accuracy
  - f1
tags:
  - dga-detection
  - streaming

DGA SGD Detector

This model was trained on the asobirov/dga-preprocessed dataset using:

HashingVectorizer (char 2–5 grams, 2**20 hash buckets)
SGDClassifier with online partial_fit and incremental learning-rate tuning
Lightweight stats: length, digit ratio, Shannon entropy
Streaming training on ~5 M domains, evaluated on ~1 M test domains from harpomaxx/dga-detection and YangYang-Research/dga-detection
Decision threshold optimized for F1 (=0.0794) yielding 85.7% accuracy, 83.6% DGA recall