Table 8.
Experimental results obtained with deep and BRISK based handcrafted features and different ML methods.
| Model & Feature extraction | Final layer | Avg Acc (%) | Avg Sens (%) | Avg Spec (%) | Avg Prec (%) | Avg F1-score (%) |
|---|---|---|---|---|---|---|
| BRISK VGG-16 | Softmax | 95.9 | 93.9 | 96.9 | 93.9 | 93.9 |
| SVM | 95.9 | 93.9 | 96.9 | 93.9 | 93.9 | |
| RF | 95.9 | 93.8 | 96.9 | 93.9 | 93.9 | |
| XG-Boost | 95.6 | 93.6 | 96.8 | 93.7 | 93.5 | |
| BRISK VGG-19 | Softmax | 96.5 | 94.6 | 97.2 | 94.6 | 94.6 |
| SVM | 96.4 | 94.6 | 97.2 | 94.6 | 94.6 | |
| RF | 96.6 | 95.0 | 97.4 | 95.0 | 95.0 | |
| XG-Boost | 96.1 | 94.4 | 97.1 | 94.2 | 94.3 | |
| BRISK sCNN | Softmax | 94.7 | 92.3 | 96.0 | 92.1 | 92.1 |
| SVM | 95.2 | 93.0 | 96.4 | 92.8 | 92.8 | |
| RF | 95.4 | 93.3 | 96.5 | 93.1 | 93.2 | |
| XG-Boost | 95.4 | 93.3 | 96.5 | 93.1 | 93.2 | |
| BRISK DenseNet 169 | Softmax | 95.2 | 92.9 | 96.5 | 93.0 | 92.7 |
| SVM | 94.9 | 92.5 | 96.2 | 92.5 | 92.5 | |
| RF | 95.2 | 92.8 | 96.4 | 93.0 | 92.8 | |
| XG-Boost | 95.2 | 92.9 | 96.5 | 93.0 | 92.7 | |
| BRISK DenseNet 201 | Softmax | 94.9 | 92.4 | 96.2 | 92.5 | 92.4 |
| SVM | 94.7 | 92.2 | 96.0 | 92.1 | 92.1 | |
| RF | 94.5 | 91.8 | 95.8 | 91.8 | 91.8 | |
| XG-Boost | 94.5 | 91.8 | 95.8 | 91.8 | 91.8 |