Research On ARIMA-LSTM In Prediction Of AIDS

AIDS(AIDS)has a high infectivity and fatality rate,bringing serious psychological and economic pressure on patients,and now it is a public health problem strongly valued by countries around the world.Therefore,understanding the general epidemic characteristics and trends of AIDS,establishing relevant prediction models can effectively predict the number of cases of AIDS,which will contribute to the reasonable allocation of medical resources and prevention and control work,and can also provide analytical ideas and methods for other infectious diseases.The development trend of AIDS is both linear and non-linear.This thesis collects the original data of the number of AIDS from 2009 to 2022 in the notifiable infectious diseases report of the Chinese Center for Disease Control and Prevention as the development trend,takes the data from 2009 to 2021 as the training set,and the data from the whole year in 2022 as the test set.In this thesis,we first establish an ARIMA model for the training set to fit the linear trend in the sequence,and select the relative best model according to the AIC / BIC criteria,and verify the model and fit in the test set to observe its prediction effect.Secondly,unlike the previous BP modeling prediction of the number of AIDS,considering the LSTM has its powerful memory function,can use a long sequence information to establish a learning model,and then find the data in the time series data interdependence,automatically detect the best mode suitable for related data,so this paper use deep learning LSTM model,used to fit the nonlinear trend in the sequence,and in the process of modeling using the Bayesian optimization algorithm for super parameters,find the best parameters of the model,the optimal parameters of LSTM model into the test set to verify the fitting.Finally,considering that the two single models have different effects,in order to combine the advantages of the two,the two are intelligently combined and estimated in parallel and series respectively.In the example modeling analysis,we established five models: ARIMA model,BO-LSTM model,equal weight method parallel ARIMA-LSTM,superiority matrix method parallel ARIMA-LSTM and series ARIMA-LSTM model,with average absolute error(MAE),root mean square error(RMSE),absolute percentage error(MAPE)as the selection basis,and finally selected the relatively best prediction model for the number of AIDS cases.The results showed that the MAE = 501.5,RMSE = 617.26,and MAPE = 0.107.the model had the highest fitting accuracy,which was the best model,which can be used to predict the trend of AIDS incidence.