Table 4 Simulation parameters for IDS Evaluation.
Parameter | Description |
|---|---|
Hardware environment | |
Server | Intel Xeon 2.4 GHz CPU, 32GB RAM, NVIDIA RTX 3060 GPU |
Emulated IIoT Devices | 50 Raspberry Pi 4 (4GB RAM) |
Dataset configuration | |
Datasets | Edge-IIoTset, IoT-23 ,MQTTset |
Data split | 70% training, 15% validation, 15% testing |
Feature selection | Mutual Information (MI), reducing feature space by 40% |
Image conversion (for baseline) | 64 × 64 grayscale images, bilinear interpolation |
Model hyperparameters | |
GNN architecture | 2 graph convolutional layers, mean aggregation, ReLU activation, softmax output |
GNN learning rate | 0.001 (Adam optimizer) |
GNN batch size | 32 |
EfficientNetV3-SVM | Pre-trained EfficientNetV3, fine-tuned on IIoT data, SVM with hinge loss |
LSTM architecture | 2 LSTM layers, 128 hidden units, dropout 0.2, softmax output |
IDS-MTran architecture | Multi-scale Transformer, 4 attention heads, 256 hidden units |
CNN-LSTM + PSO architecture | 3 CNN layers, 2 LSTM layers, PSO for feature selection, softmax output |
HE | |
Encryption scheme | CKKS (Microsoft SEAL 4.0) |
Polynomial degree | 8192 |
Encryption/Decryption time | 2.5–2.7 ms/sample (optimized) |
Evaluation metrics | |
Metrics | Accuracy, Precision, Recall, F1-Score, False Positive Rate (FPR), Processing Time (s/sample), Memory Usage (MB) |
