PREDICTION OF SHORT OR LONG LENGTH OF STAY COVID-19 BY MACHINE LEARNING

Purpose: To predict with high accuracy, using machine learning, if the length of stay Covid-19 patients will be short or not, based on basic clinical parameters.
 
 Method: Seven primary variables, including age, gender, stay, c-reactive protein, ferritin, lymphocyte, and COVID-19 Reporting Data System (CORADS), were scanned analyzed for 118 adult hospitalized diagnosis between November 2020 January 2021. Randomly, data set is split into a training 80% test 20%. Using caret package R programming language, learning models attempted to whether was long, performance values these recorded.
 Results: Among models, k-nearest neighbor model produced best results. According this model, following estimation results hospitalizations 5 days less more than days: The accuracy rate 0.92 (95% CI, 0.73-0.99), no-information 0.67, Kappa 0.82, F1 score 0.89 (p=0.0048).
 Conclusion: By applying Covid-19, estimations may made accurately, providing effective patient management.