Table 5 Hyperparameter settings and descriptions.
From: An effective brain stroke diagnosis strategy based on feature extraction and hybrid classifier
Hyperparameter | Value | Meaning |
|---|---|---|
Optimizer | Adam | Adaptive optimizer for efficient training |
Learning rate | 1 × 10−4 | Controls update step size |
Batch size | 32 | Samples per training step |
Epochs | 8 | Full passes over the dataset |
Scheduler | StepLR (step = 7, γ = 0.1) | Reduces learning rate after 7 epochs |
Loss function | CrossEntropyLoss | Used for classification tasks |
Dropout (VGG16) | 0.5 | Prevents overfitting by dropping neurons |
Input image size | 224 × 224 | Standard input size for models |
ViT patch size | 16 × 16 | Patch size for Vision Transformer |
Feature vector size | VGG16: 4096, ViT: 768 | Output dimensions from each model |
Fusion method | Feature Concatenation | Combines features from both models |
GPU used | NVIDIA Tesla P100 | Hardware used for training |
