Prediction Of Acute Kidney Injury And Death Risk Based On Machine Learning

Acute kidney injury is a clinically common syndrome,its heterogeneous etiology and pathophysiology complicates diagnosis and medical management,in addition,it may increase the poor prognosis of patients and generate a lot of medical expenses,which will increase the economic burden of patients.In actual medical diagnosis,due to the reliance on the measurement of serum creatinine,the diagnosis of acute kidney injury may be delayed and may have irreversible consequences,so early detection of acute kidney injury is crucial to improve patient outcomes.Recently,the application of machine learning methods in the medical field has become a trend.Using electronic health record data of patients can discover how these medical characteristics affect the prognosis of acute kidney injury and provide supporting information for disease treatment.This paper mainly applies Stacking ensemble learning to build a risk prediction model for acute kidney injury,uses the deep forest framework to build a mortality risk prediction model for acute kidney injury,and compares it with traditional machine learning models such as logistic regression,random forest,and decision tree.The samples selected in the empirical analysis part are from the MIMIC-Ⅲ database.Demographic information,physiological indicators,commonly used laboratory test data,comorbidities and other characteristics were mainly selected as research indicators.I selectively delete and fill data for missing values in the data.Due to the complexity of the data sample features,the regularized linear support vector machine and multicollinearity methods were used to preliminarily screen the data features in the two experiments,and XGBoost was used to analyze the data features in the two experiments.Dimension reduction processing to obtain the feature variables that are finally used for model prediction.In order to confirm the stability and generalization ability of the prediction model,different prediction models were analyzed using model evaluation indicators such as AUC,sensitivity,specificity,and accuracy.The model results show that for the classification and prediction of the risk of acute kidney injury,the stacking ensemble learning is the best;for the prediction of the risk of death,the results based on the deep forest framework are better.In this paper,the related research on acute kidney injury has achieved good results in the risk of acute kidney injury and the prediction of death.It provides data support for relevant medical decision-making and has important reference significance for the treatment of patients.