Research On Medical Expenses Prediction Based On Bayesian Network And Regression Analysis Fusion Algorithm

With the advancement of China’s medical system reform and the improvement of medical consumption level,problems such as unreasonable allocation of medical resources,unequal information of patients and doctors,and opaque medical information are gradually exposed.Therefore,the study of medical cost prediction after diagnosis can not only provide data support for the allocation of medical resources according to the condition,but also provide objective cost prediction for patients,and assist patients in the selection of treatment options.This thesis proposes a medical cost prediction method based on the fusion of Bayesian network and regression method.Firstly,the electronic medical records of patients are processed numerically and the missing values are interpolated.Then,the Bayesian classification network of maximizing conditional likelihood function and the elastic net regression method of data distribution weighted iterative parameters are combined to establish a model that can predict the treatment plan and treatment cost of patients.Finally,the influence of doctors on treatment cost is analyzed.And the doctors are evaluated by entropy method and analytic hierarchy process.And the prediction model of treatment cost is improved by using the evaluation results.The work and innovation of this thesis are as follows:(1)In order to solve the problem of high proportion of case text data,this thesis proposes a numerical method of text case information,which classifies the extracted features according to the severity and carries out numerical transformation.The clustering interpolation method is improved by distance weighting.The error of missing value interpolation is reduced(2)Aiming at the problem of insufficient prediction accuracy of traditional regression methods,this thesis first proposes a method of classification before regression,and then maximizes the conditional log likelihood function of Bayesian network to improve the classification accuracy.At the same time,according to the data distribution density,the regression iteration weight is allocated to reduce the cost prediction error.Finally,the classification probability of Bayesian network is regarded as the weight,Through weighted optimization of various regression models,The accuracy of prediction is obviously better than that before improvement.(3)At present,most of doctors’ evaluation information comes from patients’ subjective evaluation and lacks objective data support.In this thesis,the entropy method and analytic hierarchy process are combined to design an evaluation mechanism for doctors in terms of treatment costs,and the evaluation results are used as the influence variables of different doctors to optimize the Bayesian network and regression method fusion model.The accuracy of the model is improved.(4)According to the actual demand of medical cost prediction,this thesis designs and implements a medical cost prediction system by using the fusion algorithm of Bayesian network and regression analysis.The system realizes the functions of user login,case information input,treatment scheme and cost prediction,and meets the demand of medical cost prediction.