README.md

Attention Temporal Graph Convolutional Network for Cyber Physical Attacks Detection

Supervised Model: Attention Temporal Graph Convolutional Networks.

1. Attacks Detection Scheme

2. Reproducibility

- tensorflow == 1.14 (conda install -c conda-forge tensorflow=1.14)
- python == 3.7 
- scipy (conda install -c anaconda scipy)
- numpy (conda install -c anaconda numpy)
- matplotlib (conda install -c conda-forge matplotlib)
- pandas (conda install -c anaconda pandas)
- math
- sklearn (conda install -c anaconda scikit-learn)

3. A3T-GCN Architecture

4. Results

BATADAL Ranking

Author	Number of Attacks Detected	S	S_TTD	S_CM	TPR	TNR
Housh and Ohar	7	0.97	0.965	0.975	0.953	0.997
Abokifa et al	7	0.949	0.958	0.944	0.921	0.959
HCAE	7	0.933	0.947	0.918	0.865	0.972
Tsiami et al	7	0.931	0.934	0.928	0.885	0.971
Giacomoni et al	7	0.927	0.936	0.917	0.838	0.997
Brentan et al	6	0.894	0.857	0.931	0.889	0.973
A3T-GCN	7	0.845	0.839	0.851	0.774	0.927
Chandy et al	7	0.802	0.835	0.768	0.857	0.678
Pasha et al	7	0.773	0.885	0.66	0.329	0.992
Aghashahi et al	3	0.534	0.429	0.64	0.396	0.884

Forecasting Performance on Normal Dataset

	Baseline Model	Robust Mahalanobis Distance, Attention
Minimum RMSE	6.858166163	5.960369429
Minimum MAE	3.3477044	2.7673762
Maximum Accuracy	0.8372700512	0.8585200906
R2	-0.6772449017	-0.6805173159
Variance	0.9530872479	0.9646917097

• Result.

Attack Detection

	Baseline Model	Robust Mahalanobis Distance, Attention
Precision	0.6355932203	0.7208237986
Recall / True Positive Rate	0.5528255528	0.773955774
F1 Score	0.5913272011	0.7464454976
Accuracy	0.8496131528	0.8965183752
Specificity / True Negative	0.9223359422	0.9265502709

Attack Localization

Attacks Labels	Attacks Description	Feature Localization of A3T-GCN
Attack 8	Alteration of L_T3 thresholds leading to underflow	P_J256 = 11, L_T3 = 3, P_J289 = 2, L_T2 = 2
Attack 9	Alteration of L_T2	P_J289 = 13, P_J422 = 13, P_J300 = 5, L_T7 = 2
Attack 10	Activation of PU3	F_PU3 = 38, P_J280 = 28, L_T7 = 23, L_T4 = 6, P_J269 = 6, F_PU1 = 8, F_PU9 = 2
Attack 11	Activation of PU3	F_PU3 = 36, P_J280 = 31, L_T7 = 23, F_PU1 = 22, L_T4 = 12, L_T6 = 11, P_J307 = 7, P_J415 = 3, F_PU6 = 2, P_J289 = 2
Attack 12	Alteration of L_T2 readings leading to overflow	P_J289 = 7, P_J300 = 6, L_T2 = 2
Attack 13	Change the L_T7 thresholds	L_T6 = 2
Attacls 17	Alteration of T4 signal	L_T4 = 8, L_T7 = 5, P_J415 = 4, L_T6 = 2

5. References

• TGCN: A Temporal Graph Convolutional Network for Traffic Prediction
• A3T_GCN: Attention Temporal Graph Convolutional Network for Traffic Forecasting