Table 4.
The overall classification scores of proposed hybrid deep learning models and the other existing pre-trained CNN model for COVID-19 Radiography database [35].
| Deep learning models | Accuracy | Precision | recall | Specificity | Sensitivity | f1-score | MCC | JI | CSI |
|---|---|---|---|---|---|---|---|---|---|
| Vgg16 | 83% | 88% | 83% | 60.71% | 97.56% | 84% | 0.655 | 0.761 | 0.855 |
| Vgg19 | 78% | 86% | 78% | 54.92% | 96.59% | 80% | 0.582 | 0.713 | 0.782 |
| EfficientNetB0 | 76% | 84% | 76% | 51.97% | 95.76% | 77% | 0.540 | 0.676 | 0.754 |
| ResNet50 | 78% | 87% | 78% | 55.2% | 97.72% | 80% | 0.600 | 0.715 | 0.787 |
| Hybrid deep learning model (max pooling layer)-NN | 91% | 93% | 91% | 73.96% | 99.39% | 91% | 0.796 | 0.873 | 0.906 |
| Hybrid deep learning model(average pooling layer)- Naive Bayes | 63% | 76% | 63% | 79.45% | 57.97% | 66% | 0.328 | 0.541 | 0.636 |
| Hybrid deep learning model(average pooling layer)-Random Forest | 73% | 84% | 73% | 92.97% | 64.67% | 74% | 0.507 | 0.631 | 0.721 |
| Hybrid deep learning model(average pooling layer)-KNN | 68% | 82 % | 68% | 91.03% | 59.40% | 70% | 0.446 | 0.575 | 0.678 |
| Hybrid deep learning model(average pooling layer)-SVM (rbf) | 87% | 91% | 87% | 97.08% | 83.33% | 88% | 0.729 | 0.824 | 0.869 |
| Hybrid deep learning model(average pooling layer)-SVM (sigmoid) | 83% | 88% | 83% | 94.18% | 79.20% | 84% | 0.659 | 0.776 | 0.834 |
| Hybrid deep learning model(average pooling layer)-SVM (linear) | 92% | 93% | 92% | 96.29% | 89.55% | 92% | 0.805 | 0.883 | 0.913 |
| Hybrid deep learning model (average pooling layer)-NN | 92% | 93% | 92% | 77.77% | 98.68% | 92% | 0.814 | 0.889 | 0.917 |