A Readmission Risk Prediction Method For Chronic Disease Patients Based On Ensemble Learning

Chronic diseases are one of the leading causes of death in the world,threatening people’s health and lives,and imposing a heavy burden on the economic development of society.In China,chronic diseases are also a major cause of affecting national health.Due to the characteristics of chronic diseases that are difficult to cure and have a high incidence rate after discharge,therefore,when to discharge chronic disease patients is a challenging and complex process.Inappropriate discharge decisions may lead to readmission of patients,which will not only affect the physical and mental health of patients,but also cause waste of medical resources.Therefore,many medical service institutions set the readmission rate of chronic disease patients as an important indicator to evaluate the quality and utilization of medical resources.By assessing the possibility of patients being readmitted for the same or related diseases within a certain period of time after discharge,preventing avoidable readmissions can help medical staff develop personalized discharge plans and follow-up arrangements,improve patients’ quality of life,reduce medical costs and resource waste,improve the quality and efficiency of medical services,and also provide reference for the formulation and evaluation of medical policies,promote the reform and optimization of the health care system.With the development of big data,information technology is gradually applied to various fields.Due to the characteristics of data diversity and breadth in the field of health care,machine learning technology can help medical staff discover hidden patterns,trends and opportunities in data,thereby improving the efficiency and quality of decision-making and solving various complex problems.As one of the main research contents of machine learning at present,ensemble learning can improve the generalization ability and stability of models.In order to further improve the accuracy of readmission risk prediction for chronic patients.This paper proposes a readmission risk prediction method for chronic disease patients based on ensemble learning.A prediction method that uses genetic algorithm and deep Q network to dynamically select decision tree ensemble is constructed.This method first randomly generates a large number of diverse decision trees using personal basic information and clinical symptom diagnosis features in electronic medical record data,then uses genetic algorithm to select these decision trees to select the more effective decision trees for prediction.Then use deep Q network to dynamically select again which decision tree models need to participate in the ensemble according to the patient’s features.Finally,soft voting is used to obtain the final prediction result.In order to test the effectiveness of the proposed method.This paper takes a hospital in Sichuan Province as a case study,first sorts out its two-year chronic disease patient hospitalization record data,and cleans out data sets that can be used for experiments through data preprocessing.Subsequently,comparative experiments are designed to compare the performance of the proposed prediction classification method and single model and related ensemble learning on predicting 30-day readmission risk for chronic disease patients.The experimental results show that on one hand,the proposed ensemble method outperforms decision tree,logistic regression,naive Bayes and other models in terms of accuracy,recall rate,F1 score and AUC indicators.Compared with other ensemble algorithms,it also has some advantages.On the other hand,it can help medical staff provide basis for making reasonable discharge plans for chronic disease patients,reduce their readmission risk,thereby improving medical service treatment and efficiency.