A FRAMEWORK FOR EFFECTIVE PREDICTION AND CLASSIFICATION OF CORONA VIRUS (COVID-19) DISEASE BASED ON DIFFERENT CONVOLUTIONAL NEURAL NETWORK

Abstract

Corona virus (Covid-19) is one of the severe diseases that affects pneumonia and impacts our different body parts. This virus was origin from Wuhan city of China in December 2019 and later, it became a global pandemic disease rapidly spreading all over the world. To prevent viral spread, positive cases must be identified early and infected patients must be treated as soon as possible. The demand for COVID-19 testing kits has grown, and many developing countries are running out of them as new cases emerge every day. In this case, current research using radiology imaging (such as X-ray and CT scan) has been shown to be effective in detecting COVID-19, as CT scan images offer vital information about the disease caused by the COVID-19 virus.. In order to handle these problems, we have proposed different CNN models (like traditional Convolutional, Dilated Convolution and Separable Convolution) for the accurate and rapid prediction of the disease, assisting in mitigating the problem of scarcity of testing kits. The architecture consist of three convolution layers of 32 filters with kernel size of 3x3, pooling size of 2x2 and fully connected layer 1024. For performance assessment, 11,000 CT Scan images has been collected from COVID-19 CT-Scan dataset and performed three k-fold validation (3-fold, 5-fold and 10-fold) process. Experimental results shows that 10-fold cross validation model for the Covid-19 disease classification outperformed the other two k-fold cross validation (3-fold and 5-fold) by achieving 94.85%, 96.85% and 97.18% of accuracy respectively.