Covid-19 Diagnosis Model Using Deep Learning with Focal Loss Technique

Abstract

Coronavirus is an extreme virus, which spreads by human contact, now affects more than two hundred countries across the world. In comparison, new coronavirus signs are very close to the general seasonal influenza. The screening of infected people in the war against COVID-19 is seen as a crucial move. Since the positive case prediction tools of COVID-19 are not widely usable, the need for diagnostic support tools has increased. It is also of high priority that promising cases are identified earlier as possible to guarantee that this disease does not spread further. In this study, a deep learning model has been designed to diagnose Covid-19 with focal loss technique to overcome the imbalanced dataset. The results of these models have been evaluated using accuracy, recall, precision, and F1 score. The best performance achieved using the focal loss technique reached an accuracy of 89.41%, a recall of 92.6%, and a precision of 86.62%.