Table 2.
Summary of model architectures, learning parameters, and training configurations
| Model | Architecture Details | Learning Parameters | Cross-Validation (K-fold = 5) | Activation Function |
|---|---|---|---|---|
| ResNet50 | Input shape: (128, 128, 3), Input Layer: ImageNet weights, GlobalAveragePooling2D, Dense: 256, Frozen Layers: All except last block | Optimizer: Adam, Learning Rate: 0.0001, Loss: Binary Cross-Entropy, Early Stopping Patience: 15, Batch Size: 8, Epochs: 70 | Yes | ReLU (hidden), Sigmoid (output) |
| Xception | Input shape: (128, 128, 3), Input Layer: ImageNet weights, GlobalAveragePooling2D, Dense: 256, Frozen Layers: All except last block | Optimizer: Adam, Learning Rate: 0.0001, Loss: Binary Cross-Entropy, Early Stopping Patience: 15, Batch Size: 8, Epochs: 70 | Yes | ReLU (hidden), Sigmoid (output) |
| VGG16 | Input shape: (128, 128, 3), Input Layer: ImageNet weights, GlobalAveragePooling2D, Dense: 256, Frozen Layers: All except last block | Optimizer: Adam, Learning Rate: 0.0001, Loss: Binary Cross-Entropy, Early Stopping Patience: 15, Batch Size: 8, Epochs: 70 | Yes | ReLU (hidden), Sigmoid (output) |
| CNN | Input: (128, 128, 1) → Conv2D (32 filters, 3 × 3, ReLU) → MaxPooling2D (2 × 2) → Dropout (0.3) → Conv2D (64 filters, 3 × 3, ReLU) → MaxPooling2D (2 × 2) → Dropout (0.3) → Conv2D (128 filters, 3 × 3, ReLU) → MaxPooling2D (2 × 2) → Dropout (0.3) → Conv2D (256 filters, 3 × 3, ReLU) → MaxPooling2D (2 × 2) → Dropout (0.3) → Conv2D (512 filters, 3 × 3, ReLU) → MaxPooling2D (2 × 2) → Dropout (0.3) → Flatten → Dense (256 neurons, ReLU) → Dropout (0.5) → Dense (1 neuron, Sigmoid) | Optimizer: Adam, Loss: Binary Cross-Entropy, Batch Size: 16, Epochs: 70, Dropout: 0.3 and 0.5, Batch Norm: True | Yes | ReLU (hidden), Sigmoid (output) |
| SVM | RBF kernel classifier | Standardization: Applied, Solver: Sequential Minimal Optimization (SMO), Cross-validation: 5 folds, Kernel: RBF, Regularization parameter C = 1.0, Gamma: Scale, Probability estimates: Enabled | Yes | Sigmoid (output) |
| DT | Tree-based with Gini impurity | Standardization: Applied, Cross-validation: 5 folds, Splitting criterion: Gini impurity, Maximum depth: Unlimited, Minimum samples per split: 2 | Yes | - |
| RF | Ensemble of 100 trees (bootstrapped) | Standardization: Applied, Cross-validation: 5 folds, Total estimators: 100, Splitting criterion: Gini impurity, Maximum number of features: Auto | Yes | - |