Research On Patient Satisfaction Prediction In Online Medical Community Based On XGBoost Algorithm

The online medical community enables patients to contact doctors without time and space constraints.In addition to convenience and cost,patients can also obtain diverse medial suggestions from different doctors,which allow them to make better medical decisions.Doctors can obtain social and economic benefits from engagement in online medical communities.The increase in the number of registered patients and doctors in the online medical community has led to the problem of information overload.It is difficult for patients to choose a doctor that satisfies them among the many doctors.At the same time,there is information asymmetry between doctors and patients,and patients cannot predict whether the selected doctor is satisfied or not.Patient satisfaction of doctors can help patients pre-select their doctors.Understanding the factors that affect patient satisfaction and predicting patient satisfaction can help patients narrow their choices.The research can also help medical staff to improve the quality of medical services in a targeted manner.At present,there are more and more researches on patient satisfaction.Most researches focus on exploring the factors that affect satisfaction,but there are few researches on satisfaction prediction.In the online medical community,there are many types of information related to doctors.How to systematically use this information to predict patient satisfaction requires further research.In this paper,we crawled all the doctor information and patient information of the endocrinology department of Good Doctor Online,extracted 14 features through data preprocessing and numerical transformation which was divided into four dimensions including competence characteristics of doctors,online efforts of doctors,service evaluation of doctors,and patient medical treatment process.We retained all 14 features and used decision tree,random forest,GBDT and XGBoost algorithm to predict patient satisfaction.Due to the extreme imbalance of the patient satisfaction data,the SMOTE algorithm was used to process the data and compare the model prediction before and after using the SMOTE algorithm to further analyze the impact of data imbalance on prediction.The AUC,Kappa and confusion matrix,which are more sensitive to imbalanced data,are used to evaluate the prediction model.Finally,XGBoost is obtained as the optimal model,and the model is further trained using tenfold crossvalidation to prove the effectiveness of the prediction model.The top four ranked features were analyzed exploratively by ranking the features using the importance of the features(based on average information gain)generated by the XGBoost algorithm.Doctor's active degree(online efforts of doctors)is positively related to patient satisfaction.The treatment method received by the patient(patient medical treatment process)affects patient satisfaction.The total number of patients treated by the doctor(service evaluation of doctors)is positively related to patient satisfaction.There is no clear positive or negative relationship between high or low acdemic title(competence of doctors)on patient satisfaction,but patients were less satisfied with professors than with associate professors.In addition,other features such as registration way,the reasons why patients choose doctors,medical expenses,region,hospital department,degree and clinical title can also influence patients' satisfaction.To investigate the effect of feature combination on the results of satisfaction prediction,the top six features generated by XGBoost algorithm in terms of feature importance were used to model and predict with features combined one by one.The results demonstrates that more features may generate better and more stable results.