Predictive value for prognosis of sepsis based on the Light Gradient Boosting machine algorithm model

Abstract

Abstract Sepsis is one of the leading causes of death in the critical care unit. The latest data that over 19 million patients every year in the world suffer from severe sepsis indicates it of great significance to evaluate the development tendency of sepsis and to investigate the prediction value of prognosis. Based on the Light Gradient Boosting (LGB) machine learning algorithm, we have now developed and tested an LGB prediction model by using the data source from the Medical Information Mart for Intensive Care-IV database for the model construction and validation, thus to predict the prognosis of sepsis. Additionally, we established three more models including the Logistic regression, the Random Forest and the K-Nearest Neighbor based prediction model and made comprehensive comparison with the LGB prediction model in the indicators involving 8 aspects, obtaining an area under the curve (AUC) of LGB prediction model at 0.998, which demonstrates it of strong reliability to exhibit high accuracy for predicting the prognosis of sepsis patients. Our findings support the LGB prediction model as a preferred machine learning model for predicting the prognosis of patients with sepsis.