Please enable JavaScript.
Coggle requires JavaScript to display documents.
DNP3 SCADA: ML vs Bayesian Roadmap - Coggle Diagram
DNP3 SCADA: ML vs Bayesian Roadmap
1) Dataset + task definition
Download DNP3 dataset (IEEE DataPort)
Define labels: Normal=0, Attack=1
2) Data preparation
Select CSVs (balanced train/test preferred)
Check class balance + missing values
Choose feature set: TCP / DNP3 / Combined
3) Preprocessing + split
Drop non-features (IDs, timestamps if risky)
Scale numeric features
Create train / validation / test (avoid leakage)
4) Train ML baselines
Logistic Regression
Random Forest
SVM or MLP
Predict test probabilities
5) Train Bayesian model (PyMC)
Bayesian Logistic Regression
NUTS sampling (posterior inference)
SMC sampling (for model evidence / Bayes factors, if needed)
Posterior predictive test probabilities
6) Evaluate + compare
Metrics: ROC-AUC, PR-AUC, F1, Log-loss, Brier
Outputs: one results table + ROC plot
7) Conclusion + SCADA recommendations
Select best model (by metrics + justification)
3 security recommendations (thresholds, alarms, retraining/drift)