Intelligent System for the Prediction of COVID-19 Infection Based on Frequency Ratio Model and Radial Base Function Neural Network

Coronavirus (COVID-19) has been affecting more than 213 countries worldwide since December, 2019. With more than 3.42 million deaths and more than 165 million infections. COVID-19 also is blamed for billions of dollars as an economic loss. It also has placed a huge burden on the healthcare system worldwide. Like other disease, COVID-19 initially was diagnosed by symptoms and later using laboratory tests. In some developing countries, symptoms are the only available method for diagnosis. In this study, Radial Base Function (RBF) neural network and Frequency Ratio (FR) were employed to classify COVID-19 data set and find the factors that are highly correlated with COVID-19 infection. Data set was obtained from ministry of health in Mexico. It includes 10 factors (attributes) age, sex, obesity, asthma, hypertension, CKDs, CVDs, tobacco, diabetes and pneumonia. 10-cross validation methods were employed in training RBF. RBF was initially trained using the 10 factors and later trained using the important factors that are chosen based on FR. The performance of RBF was evaluated using specificity, sensitivity and accuracy. FR model shows that only 7 factors have a strong relation with COVID-19: Age, pneumonia, Diabetes, CVDs, hypertension, asthma, sex and CKDs. The simulation results showed that RBF trained with selected factors have higher specificity, sensitivity and accuracy in comparison with using all 10 factors. In addition, RBF is compared with 5 different well-known methods.