Which features help the model?
Toggle features on and off. The model retrains in real time — watch which features make the biggest difference.
Raw columns
Engineered features
Accuracy
—
Precision
—
Recall
—
F1 Score
—
Comparison
Loading...
Loading...
Loading...
Think Deeper
Try this:
Remove all engineered features and train with just raw columns. Then add bytes_per_second. What happens?
Raw columns alone give decent accuracy (they're already numeric), but adding bytes_per_second often boosts recall significantly. It captures transfer speed — a derived signal that neither bytes_sent nor duration alone express.
Cybersecurity tie-in: In SOC ML pipelines, adding bytes_per_second and
port_risk_score often matters more than tuning the model algorithm. Feature engineering
is where domain expertise meets machine learning — and where security engineers have the advantage.