Conclusion

• We read the dataset and resized it to 32x32x3
• This datset was then used to train a custom Convolutional Neural Network and VGG16
• With custom convolutional neural network we got the avg. training accuracy as 98.93% and avg. training loss as 0.0298, avg. validation accuracy as 96.03%, and the testing accuracy was achived as 99.36%
• With VGG16 we got the avg. training accuracy as 98.41% and avg. training loss as 0.0432, avg. validation accuracy as 99.09%, and the testing accuracy was achived as 99.42%
• On unseen dataset which had COVID 19 CT scan images, our custom convolutional neural network performed better than VGG16. Our model gave an accuracy of 78.68% and VGG16 gave an accuracy of 46.35%

Future Scope

• Finding out the reasons for the change in performance of VGG16
• Using other pretrained convolutional neural networks like ResNet, Xception, Inception, and LeNet
• Trying machine learning models on the dataset, after extracting features from the images