Table 4.
Performance comparison of the hyperparameter tuned ablation models.
| Batch Size |
Learning Rate |
Dropout Regulari-zation |
Optimizer | Random Forest |
Val Accuracy | Test Accuracy | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hyperparameters |
Best Hyperparameter |
||||||||||||
| n_ estimators |
max_ depth |
min_ samples_ split |
n_ estimators |
max_ depth |
min_ samples_ split |
||||||||
| VGG16 + RF | 32 | 0.001 | 0.5 | ADAM | [50, 100, 200] | [None, 10, 20] | [2,5,10] | 200 | 20 | 2 | 97.5 % | 96.02 % | |
| VGG19+ RF | 200 | None | 2 | 96.89 % | 95.82 % | ||||||||
| Mobilenet + RF | 200 | None | 5 | 97.29 % | 96.63 % | ||||||||
| Resnet50 + RF | 200 | 20 | 2 | 97.36 % | 97.03 % | ||||||||
| Alexnet + RF | 50 | 9 | 5 | 63.06 % | 61.54 % | ||||||||
| Support Vector Machine | Val Accuracy | Test Accuracy | |||||||||||
| Hyperparameters | Best Hyperparameter | ||||||||||||
| C | Kernel | C | Kernel | ||||||||||
| VGG16 + SVM | 0.1, 1, 10 | Linear, RBF | 0.1 | Linear | 98.71 % | 97.84 % | |||||||
| VGG19+ SVM | 10 | RBF | 98.44 % | 97.23 % | |||||||||
| Mobilenet + SVM | 10 | RBF | 99.53 % | 98.92 % | |||||||||
| Resnet50 + SVM | 10 | RBF | 99.32 % | 98.79 % | |||||||||
| Alexnet + SVM | 10 | RBF | 47.09 % | 47.44 % | |||||||||