
SDN CAPSTONE - FINAL RESULTS FOR GUIDE
======================================

01_ORIGINAL_DATA_AND_REPORTS
All original research CSVs, JSON reports, audits, and figures.

02_NEW_H14_MODEL_RESULTS
Newly computed H14 machine-learning metrics, feature
importance values, confusion matrices and dataset analysis.

03_GRAPHS_PNG
Presentation-ready graphs.

04_PAPER_REFERENCE_TABLES
Reported H23 and H28 paper results.

PROJECT:
Dynamic and Phase-Aware SDN Dataset Generation

MAIN TOOLS:
Mininet, Ryu, Open vSwitch, OpenFlow 1.3

ML MODELS:
1. Logistic Regression
2. Random Forest
3. RBF-SVM
4. Gradient Boosting
5. Soft Voting Ensemble

DATASET:
H14: 216 observations
Training: 72
Validation: 72
Testing: 72

Label 0 = not synthetic probing
Label 1 = synthetic probing

PAPER H23 RESULTS:
Gradient Boosting balanced accuracy: 89.58%
Random Forest recall: 87.50%

PAPER H28 RESULTS:
Stable RTT: 6.05 ms
Constrained RTT: 32.88 ms
Recovery RTT: 6.28 ms

Stable TCP goodput: 9.30 Mbps
Constrained TCP goodput: 1.90 Mbps
Recovery TCP goodput: 9.26 Mbps

IMPORTANT:
New H14 model scores are from the newly trained Colab models.
They are not the original H23 frozen-model results.

The paper evaluates controlled synthetic probing activity,
not independently verified real-world malicious attacks.
