Table 4.
Performance comparison of Softmax probabilistic-based and deep feature extracted from custom-made TL-B CNNs with SVM-based classification of four best-performing TL-B CNN models selected for proposed DFS-BTD framework. 60:40% data portioning (training: testing).
| Model | DFS-HL Scheme | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Transfer Learning-Based (TL-B) Softmax Based Classification |
4 Best Performing Transfer Learning-Based (TL-B) with SVM |
|||||||||
| Acc. % | Rec. | Pre. | F1-Score | MCC | Acc. % | Rec. | Pre. | F1-Score | MCC | |
| Inception-V3 | 98.52 | 0.9924 | 0.9806 | 0.9856 | 0.973 | 99.01 | 0.9824 | 0.9950 | 0.9887 | 0.9776 |
| Resnet-18 | 98.91 | 0.9774 | 0.9966 | 0.9869 | 0.9744 | 99.16 | 0.9799 | 0.9991 | 0.9894 | 0.9793 |
| GoogleNet | 98.52 | 0.9924 | 0.9806 | 0.9856 | 0.9731 | 99.11 | 0.9849 | 0.995 | 0.9899 | 0.9801 |
| DenseNet-201 | 98.86 | 0.9724 | 0.9991 | 0.9856 | 0.9720 | 99.06 | 0.9887 | 0.9918 | 0.9902 | 0.9806 |