Ults and not just as a proof of the notion, but in addition provides insights on no matter if a system is practically feasible in real-life circumstances or not. The efficiency evaluation of your proposed methodology is accomplished making use of distinctive evaluation measures such as accuracy, precision, recall, F1measure, and confusion matrix. All these evaluation measures are derived according to the following 4 scenarios. The experiments are performed employing randomly normalized dataset determined by the minimum quantity of photos in Viral Pneumonia class, as well as working with the actual variety of photos for each class in the dataset. Similarly, the experiments are also performed employing the freeze weights of distinctive DL models also as nonfreeze weights, where we proposed to keep the best ten layers frozen and the rest in the weights unfreezed to train them again. Table 1 shows the results of applying the several optimized deep studying algorithms; VGG19, VGG16, DenseNet, AlexNet, and GoogleNet with weights frozen and applied towards the non-normalized Melitracen Technical Information information within the dataset. Outcomes indicate that the top accuracy is D-Glucose 6-phosphate (sodium) Description achieved using DenseNet with an average value of 87.41 and 94.05 , 95.31 , and 94.67 for precision, recall, and F1-measure, respectively. The lowest accuracy is reported for the VGG19 algorithm with an average value of 82.92 .Table 1. Experimental final results of distinctive models with freeze weights and non-normalized information.Accuracy VGG19 VGG16 DenseNet AlexNet GoogleNet 82.92 84.22 87.41 84.14 83.Precision 90.40 91.13 94.05 86.97 89.Recall 94.25 98.03 95.31 99.13 96.F1 Measure 92.29 94.45 94.67 92.65 92.The experiments had been then repeated around the very same optimized DL algorithms, but this time using the nonfreeze weights with normalized data, as shown in Table 2. The accuracy in this case improved considerably, with all the greatest accuracy achieved by the VGG16 with an average value of 93.96 , a precision of 98.36 , recall of 97.96 , and F1-measure of 98.16 . The lowest accuracy is reported for the GoogleNet with an average value of 87.92 . Note that with nonfreeze weights, the accuracy elevated by 6.55 than the highest reported accuracy in Table 1. Repeating the experiments together with the nonfreeze weights on the non-normalized information is shown in Table three. Here, the larger dataset increases the accuracy by around 0.3 for VGG16. The highest accuracy was once more achieved by VGG16 with an typical value of 94.23 , precision of 98.88 , recall of 99.34 , and F1-measure of 99.11 . The lowest accuracy is once more reported using the GoogleNet, with an average value of 89.15 .Table 2. Experimental results of diverse models for the nonfreeze weights and normalized data.Accuracy VGG19 VGG16 DenseNet AlexNet GoogleNet 92.94 93.96 90.61 91.08 87.Precision 99.15 98.36 95.98 96.23 92.Recall 96.68 97.96 95.60 97.87 92.F1 Measure 97.90 98.16 95.79 97.05 92.Diagnostics 2021, 11,12 ofTable 3. Experimental final results of diverse models with nonfreeze weights and non-normalized information.Accuracy VGG19 VGG16 DenseNet AlexNet GoogleNet 93.38 94.23 92.08 91.47 89.Precision 98.97 98.88 98.52 97.69 96.Recall 98.60 99.34 98.04 98.16 97.F1 Measure 98.78 99.11 98.28 97.92 96.Using the augmented normalized dataset together with nonfreeze weights, the experiments are repeated making use of the identical DL algorithms along with the results are shown in Table 4. Once more, the outcomes indicate an increase in accuracy. Although it can be a minor raise of 0.03 , this results in a far better combination that would improve accuracy significantly as compared w.
