Table 5 Overview of key hyperparameters used for feature selection with S3O and classification with efficientnet.
Hyperparameter | Value | Justification |
|---|---|---|
Optimizer | Adam | Adaptive, robust for deep learning |
Learning rate | 0.0001 | Stable training and convergence |
Batch size | 32 | Suitable for medical images and memory constraints |
Epochs | 100 | Enough to converge without overfitting |
Loss function | Categorical Crossentropy | Multi-class classification (3 classes) |
Dropout | 0.3 | Prevents overfitting during training |
Input size | 224 × 224 × 3 | Common for EfficientNet pretrained backbone |
Number of classes | 3 | Normal, Ischemic Stroke, Hemorrhagic Stroke |
