Table 7.
Computational performance of the different TL-based deep learning models.
| Deep Learning Models | 1 ACC | 1 PR | 1 SE | 1 SP | F1-Score | 1 MCC |
|---|---|---|---|---|---|---|
| ShuffleNet-Light | 99.1% | 98.3% | 97.4% | 98.2% | 98.1% | 98% |
| ShuffleNet | 88.5% | 87.3% | 88.3% | 88.7% | 87.7% | 85% |
| SqueezeNet | 87.9% | 84.5% | 90.8% | 85.4% | 87.6% | 84% |
| ResNet18 | 89.3% | 87.1% | 93.1% | 87.9% | 90.1% | 89% |
| MobileNet | 90.8% | 90.2% | 90.0% | 91.4% | 90.1% | 88.% |
| Inception-v3 | 89.4% | 87.7% | 90.0% | 88.9% | 88.8% | 88% |
| Xception | 89.5% | 88.3% | 89.3% | 90.7% | 88.7% | 88% |
| AlexNet [17] | 88.9% | 87.6% | 88.8% | 88.9% | 87.7% | 86% |
1 SE: Sensitivity, SP: Specificity, RL: Recall, PR: Precision, ACC: Accuracy, MCC: Matthew’s correlation coefficient.