abstract |
The present invention proposes a method for predicting the outcome of patients with new coronary pneumonia based on an interpretable machine learning algorithm, including: extracting patient data of COVID-19 from a database, and dividing the patient data into an experimental group and a control group according to the patient's condition conversion group; impute the missing values of each index through random forest regression; screen the indicators input to the model, and use the screened indicators as the key risk factors for identifying the deterioration of the patient’s condition; input the key risk factors of the patient into the XGBoost model and logic Regression model; select the XGBoost model with better predictive performance, generate a combination of indicators, then use the XGBoost model to make predictions, and record the prediction results; define the early warning range of key indicators; when the patient's key risk indicators enter the early warning range, medical staff will be notified. An alarm prompt is issued; the calculation results of the algorithm and the clinical experience of doctors are combined, and a combination of two indicators consisting of 15 first-group indicators and 5 second-group indicators is proposed to predict the condition of patients with new coronary pneumonia. |