Method for finding subsets of consensus features in predicting the effectiveness of rehabilitation of patients after COVID-19

Coronavirus infection causes long-term post-Covid syndrome, which determines the need for medical rehabilitation. The use of modern machine learning technologies to predict the effectiveness of rehabilitation can personalize the process of providing assistance.Aim: To create a method for constructing a model to predict the effectiveness of patients rehabilitation who have suffered COVID-19.Material and Methods. The study included 64 patients admitted for inpatient rehabilitation after COVID-19. The average age was 56.92±9.29 years. To obtain information about the patients' health status, a physical examination, six-minute walk test (SHT), clinical and biochemical blood tests, and spirometry were performed in order to obtain information about the health status of the patients. The collected data were anonymized and transformed into a data set for the classification task, the output label of which was a binary attribute indicating the presence or absence of an improvement in the six-minute walk test result by at least 15 %. The proposed method for determining a consistent subset of features was tested by ReliefF, χ2-squared filters and the minimum redundancy and maximum information content algorithm; the binary genetic algorithm NSGA2 was used as a wrapper. The construction of preliminary machine learning models on the found subsets of features using the linear support vector machine was carried out.Results. In the process of testing the proposed method in the task of predicting the effectiveness of rehabilitation of patients after COVID-19, a subset of signs was identified that made it possible to achieve the maximum value of the concordance coefficient. The found set includes the following characteristics: gender, concomitant diseases, shortness of breath, cough, complaints about the gastrointestinal tract, six-minute walk test result, D-dimer level, assessment of shortness of breath according to the Borg scale.Conclusion. The proposed method allowed identifying the signs that are of the greatest importance for predicting the effectiveness of patients’ rehabilitation who have had COVID-19.