Improving the prediction of the severity of the condition of patients infected with the Corona virus

Abstract

Background and aim: Although more than two years have passed since the beginning of the spread of the Corona virus, there are no indications that the pandemic is about to leave, despite the massive vaccination campaigns and measures against this epidemic, as well as the availability of tests to detect infection early.

During the past two years,were multiple models of deep learning using neural networks presented to reach rapid automated models that help detect cases of COVID-19, and researchers relied on a wide range of features.

This research aims to reduce the number of these features while keeping the performance of the classifier intact and knowing the most important features that help in detecting the degree of risk of infection according to the cases of recovery and death in order to stop the transmission and spread of the disease, especially in high-risk workplaces such as work environments in health care.

Materials and methods: The following algorithms were used for feature reduction: Boruta algorithm, genetic algorithm, KNN algorithm and Ridge coefficient with neural networks, and each of them was discussed according to the number of features extracted in each method.

Results: The results showed that all the extraction algorithms agreed about the age as the most important for the increased risk of infection with COVID-19. The high temperature is a common feature with various degrees of importance among the extraction algorithms. In addition, coughing, fluid leaching and the number of neutrophils play an important role in increasing the possibility of infection with the Corona virus.

Conclusion: The least number of features that helped in detecting the severity of the patients' case of infection is three according to (the Boruta algorithm and the Ridge coefficient), while the largest number of features is five, according to both the genetic algorithm and the KNN algorithm.