Table 4.

Complete implementation specifications :

Component	Specification
Data Augmentation	None applied
Feature Extraction Layer (Hybrid)	VGG16: block5_pool (25,088-D); ResNet50: avg_pool (2,048-D); Xception: avg_pool (2,048-D)
Feature Normalization	Z-score standardization (zero mean, unit variance)
SVM Hyperparameters	Kernel: RBF; C = 1.0; γ = 0.001; Fixed a priori
Decision Tree Hyperparameters	Max depth: 10; Min samples split: 5; Criterion: Gini; Fixed a priori
Random Forest Hyperparameters	Estimators: 100; Max depth: 20; Min samples split: 2; Max features: sqrt; Fixed a priori
CNN Learning Rate	0.001 (Adam optimizer)
Weight Decay	None applied
Initialization	Glorot uniform (weights), Zero (biases)
Early Stopping	Patience: 10 epochs; Monitor: validation loss
Batch Size	16
Maximum Epochs	30
Hyperparameter Tuning	Fixed a priori without grid/random search