Smart COVID-19 Prediction System Using Neural Network

Abstract

The pandemic of coronavirus COVID-19 has created a great danger and concern for humanity. Many researchers have done different types of work in this area to provide medical services. In this paper, we proposed a smart Covid-19 diagnosis system by using a Feed Forward Backpropagation Neural Network (FFBNN) and Probabilistic Neural Network (PNN). Based on personal information from patients such as (age, gender, contact with sick person) and five symptoms (headache, fever, cough, sore throat, and shortness of breath) for this purpose we used 510 samples that are collected from different sources, and then compared to previous studies. Results of this work showed that using FFBNN has achieved highest accuracy (98.0%), sensitivity (100%), specificity (94.4%), precision (97.1%), recall (100%) and F1-score (98.52%). But PNN that has accuracy, sensitivity, specificity, precision, recall, F1-score of 90.2%, 92.7%, 87.2%, 89.47%, 92.7% and 91.07% respectively. The most relevant features to positive Covid-19 were fever, shortness of breath, and cough with correlation coefficient of 0.591, 0.495 and 0.488.