Table 10 VANET-DDoSNet++ Architecture Specification.
From: A hybrid deep learning model for detection and mitigation of DDoS attacks in VANETs
Stage | Component | Configuration |
|---|---|---|
Input Layer | Raw input features | Shape: (T, F) where T = time steps, F = feature dimension |
Conv block 1 | Conv1D + ReLU | Filters: 64, Kernel Size: 3, Padding: ‘same’, Stride: 1 |
Conv block 2 | Conv1D + ReLU | Filters: 128, Kernel Size: 3, Padding: ‘same’, Stride: 1 |
Conv block 3 | Conv1D + ReLU | Filters: 256, Kernel Size: 3, Padding: ‘same’, Stride: 1 |
Attention layer | Self-Attention / Multi-Head Attention | Heads: 4, Head Dimension: 64, Scaled Dot-Product, Positional Encoding used |
Recurrent block | Bidirectional LSTM (× 2) + ReLU | Units: 128 each direction, Dropout: 0.3, Activation: ReLU |
Residual/Dense skip | Residual connections (Conv → LSTM) | Element-wise addition before LSTM input |
Fully connected layer | Dense Layer + ReLU | Units: 64, Activation: ReLU |
Output layer | Dense Layer + Softmax | Units: 2 Classes, Activation: Softmax |
