Table 1 Training parameters of EEFN-XM-PDC-HybPS-GWO.
Parameter | Description/value |
|---|---|
Convolution layers (33 conv.layers) | The EEFN-XM architecture consists of convolutional layers with different filter sizes and strides, including a stem layer and MBConv blocks |
Hidden layer (activation function) | Swish is a self-gated activation function utilized in EEFN-XM |
Attention mechanisms (4 conv. layers) | Depthwise separable convolution and Squeeze-and- Excitation Lightweight cross-modal attention mechanism |
Feature fusion layer (3 conv.layers) | Lightweight cross-modal attention mechanism fusion technique is used to concatenate image based and gait based features |
Output layer (activation function) | SoftMax is used for classification tasks, including categorizing PD into healthy controls, early and advanced stages |
Pooling layer | Global Average Pooling |
Epochs | 200 |
Batch size | 32 |
Image dimension | 224 × 224 × 3 represents the standard input size typically used for EEFN-XM |
Learning rate | 0.001 |
Optimizers used | Hybrid Particle Swarm and Grey Wolf Optimizer (HybPS-GWO) |
Random seed | 42 |
