Research On The Prediction Model Of Patient's 30-day Readmission Based On The Gradient Boosting Decision Tree

Thirty-day readmission rate refers to the rate of return to hospital within 30 days after discharge,which can reflect the effect of the first operation of the hospital.It is an important index to measure the quality of medical service in the world at present.Its application in the index system of evaluating the quality of medical service in our country has just begun.The use of prediction algorithm to predict the readmission of patients can provide decision support for hospital administrators.Thus it can effectively reduce the readmission rate,improve the quality of medical service and reduce the cost of treatment,and help the hospital to allocate medical resources more effectively and reasonably at the same time.The causes of readmission are diverse and it is difficult to assess the risk of readmission based solely on clinical expertise.At the same time,there are two main methods for predicting readmission.One is to adopt traditional statistical methods,such as Linear regression,Logical regression,etc.The other is to simplify the characteristic parameters of prediction model by using machine learning method,using Support Vector Machine,Naive Bayes,Decision tree and other methods for readmission prediction.It can effectively solve the problem of insufficient prediction accuracy of traditional methods.The data sample of readmission patients stored in medical information system is unbalanced and the number of readmission patients is a minority in all patients.Traditional statistical methods and machine learning algorithms are used to classify and predict the balanced data,which can not effectively identify a small number of samples in the unbalanced data.Thus,the ideal classification results are not obtained.In order to improve the accurate classification prediction of 30-day readmission,this paper constructs a kind of readmission prediction model based on gradient lifting decision tree algorithm.Firstly,the medical imbalance data is processed.This paper proposes a method of sampling unbalanced data using FCM algorithm,and extracts the feature of the data based on gradient lifting model.Gradient Boosting decision tree and Bayesian model were used to optimize the hyperparameter.The classification performance of the proposed algorithm on the UCI common unbalanced data set is better than that of the common sampling methods,such as sampling SMOTE andintegrated sampling strategy.Compared with the commonly used classification algorithms,the proposed BFCM-LGB algorithm is better than the decision tree,logical regression,naive Bayesian SVM and random forest classification.Finally,the results of two public readmission data experiments were compared with the conventional algorithm.This algorithm has significant advantages in prediction accuracy and recall rate,compared with the existing classification algorithms of all-cause readmission and diabetes readmission.The results obtained by the method presented in this paper have high prediction accuracy.