CICIDS LightGBM Gatekeeper (Stage 1)
Binary classifier for the Hybrid Edge-to-Core IDS. Acts as a fast filter to distinguish benign from malicious network flows.
Performance
- Accuracy: 99.87%
- Precision: 99.69%
- Recall: 99.89%
- Macro F1: 99.85%
- Optimal Threshold: 0.4837
Architecture
- Type: Gradient Boosted Decision Trees
- Estimators: 250
- Tuning: Optuna (10 trials)
- Input: 58 numerical flow features
Usage
import lightgbm as lgb
model = lgb.Booster(model_file='stage1_gatekeeper.txt')
prob = model.predict(features.reshape(1, -1))[0]
is_attack = prob >= 0.4837