Prediction Of Medical Service Waiting-time Based On Ensemble Learning Algorithm

The prediction of service waiting time is the key premise for the service organization to implement scientific management.It can not only help to know the real-time waiting status for customers,but also help to realize real-time service guiding and to improve quality for service providers.In this paper,the prediction of medical service waiting time is selected as the research subject.According to the complexity of the problem,the prediction models of single-stage waiting time and multi-stage service waiting time are refined and applied to the waiting time prediction of medical service institutions to provide decision support for improving patient satisfaction and medical institution management.Detailed researches are carried out from the perspective of prediction models and application analysis,including:From the perspective of prediction model algorithms,GBDT,Xgboost,and Light GBM algorithms are first compared,and the Light GBM algorithm is determined as the basic prediction method.The Bayesian optimization idea is further introduced to optimize the hyperparameters which affect the efficiency of the algorithm in Light GBM.The process is then transformed into the problem of optimizing the objective function in a given optimization space,and a BO-Light GBM prediction algorithm is proposed.The algorithm is verified through multiple sets of standard machine learning data sets.The results show that the BO-Light GBM algorithm can reduce the optimization time by more than 90% compared with grid optimization.Compared with random optimization,the BOLight GBM algorithm has prediction accuracy with 1.48% ～ 9.58% increasing.At the same time,the BO-Light GBM algorithm is also found to be superior to the GBDT and Xgboost models.From the perspective of application analysis,for single-stage waiting time prediction,a process in a Ningbo Maternal and Child Health Hospital was used as the research subject to collect relevant data of the hip examination department.The current numbers of queue,the time when patients entered the queue,the current date,and the influence of weather and other influencing factors are involved in the model’s inputs.The prediction results show that the BO＿Light GBM algorithm has good prediction performance,with an average absolute error of about 2.0 minutes and an average percentage error of 21.60%,which provides a reference for practical application problems.Compared with the single-stage waiting time prediction,the multi-stage service waiting time prediction has the following differences:(1)Patients have stronger randomness,which is reflected in the multiple selectivity of patients in various departments;(2)For a single department,it is found that the numbers of queue is one of the most critical indicators that affect the waiting time of patients.In the multi-stage service waiting time prediction,this paper takes the numbers of all queues when the patient enters the first examination unit queue as a predictive feature,and also includes the total number of examination units,and features such as the time of entering the queue.GBDT,Xgboost,and Light GBM are also adopted for prediction and used for comparision,taking into account the optimization time and prediction accuracy.The BO＿Light GBM model based on Bayesian optimization algorithm has the best prediction performance,with an average percentage error of8.9%.