Abstract: The COVID-19 pandemic, caused by the SARS-CoV-2 virus, has triggered a global health crisis, necessitating accurate predictive models to forecast disease severity and aid in clinical decision-making. This study introduces an innovative machine learning approach, the bDWPLO-FKNN model, designed to predict the severity of COVID-19 pneumonia in patients. The model incorporates the Differential Weibull Polar Lights Optimizer (DWPLO), an enhancement of the Polar Lights Optimizer (PLO) with the differential evolution operator and the Weibull flight operator, to perform effective feature selection. The DWPLO’s performance was rigorously tested against IEEE CEC 2017 benchmark functions, demonstrating its robust optimization capabilities. The binary version of DWPLO (bDWPLO) was then integrated with the Fuzzy K-Nearest Neighbors (FKNN) algorithm to form the predictive model. Using a dataset from the People’s Hospital Affiliated with Ningbo University, the model was trained to identify patients at risk of developing severe pneumonia due to COVID-19. The bDWPLO-FKNN model exhibited exceptional predictive accuracy, with an accuracy of 84.036% and a specificity of 88.564%. The analysis revealed key predictors, including albumin, albumin to globulin ratio, lactate dehydrogenase, urea nitrogen, gamma-glutamyl transferase, and inorganic phosphorus, which were significantly associated with disease severity. The integration of DWPLO with FKNN not only enhances feature selection but also bolsters the model’s predictive power, providing a valuable tool for clinicians to assess patient risk and allocate healthcare resources effectively during the COVID-19 pandemic.

Key words: COVID-19 severity prediction, Polar lights optimizer, Differential evolution, Weibull flight operator, Fuzzy k-nearest neighbors')">COVID-19 severity prediction, Polar lights optimizer, Differential evolution, Weibull flight operator, Fuzzy k-nearest neighbors