A Comparative Study On Application Of ARIMA Model And BP Neural Network Model To Predict The Incidence Of Aids

Objective AIDS is widespread in our country,and there is growing trend,fatality rate is high,the disease has become our country and the world a serious public health problem and social events,which cause serious damage to people's life and health,affect all aspects of the society.Therefore,establish early warning technology and the development of HIV/AIDS,so as to find disease early development trend,is of great significance to the HIV/AIDS prevention and control work.AIDS prediction method starts late in our country,to a relatively rapid development in the 90 s.At present most of scholars used traditional linear model to predict incidence of HIV/AIDS.Also have some scholars in trying the nonlinear prediction model.Two kinds of prediction model show that the fitting effect and prediction effect is good.Chinese AIDS incidence data present linear and nonlinear characteristics.This research intends to adopt the typical traditional linear model of ARIMA model and nonlinear BP neural network model with incidence of HIV/AIDS in establishing model,and then predict incidence of HIV/AIDS,finally compare the two different model fitting prediction effection.To explore a suitable prediction model for AIDS incidence in our country,to provide a scientific and feasible disease prediction model for AIDS prevention and control works.Method Published by the ministry of health in China from 2004 to 2014 on AIDS incidence as the research object,using BP neural network model and ARIMA model respectively,we set up the a predicting model for incidence of HIV/AIDS,to find the best structure,then to predict and analysis the incidence of HIV/AIDS.Through the analysis of the mean square error(MSE),mean absolute error(MAE),mean absolute percentage error(MAPE)three indicators to compare two models'fitting effect and prediction effect,and then to evaluate and compare the two models.Results With the incidence of HIV/AIDS from 2004 to 2013 months as the original data to establish the model,to predict the incidence of AIDS in 2014.Compare forecast incidence of AIDS and actual incidence of 2014,to verify the model fitting effect.(1)The ARIMA model.Ultimately selected ARIMA(0,1,1)(0,1,1)structure for the best model of time sequence whose white.noise testing LB(18)= 14.853,P>0.05.This means that the model is effective.Fitting error parameter values of the model are:MSE = 0.168,MAE =0.026,the MAPE= 22.589;Prediction error parameter values are:MSE = 0.0006,MAE=0.190,MAPE = 0.785.(2)The BP neural network model.Ultimately selected the BP neural network model that have 3-8-1 structure and using LM algorithm as the optimal model.The model training error MSE is 0.0019,completed a total of 16 iterations.Fitting error of the model parameter values are:MSE = 0.0004,MAE = 0.0143,MAPE = 9.9072;Prediction error parameter values are:MSE = 0.0009,MAE = 0.0083,MAPE = 0.3405.Thus,the BP neural network model's fitting error is smaller than ARIMA model,forecasting accuracy is higher than the ARIMA model.Conclusion ARIMA model and BP neural network model can be used in the monthly incidence of HIV/AIDS forecasting,but the nonlinear BP neural network model fitting and forecast effect is better than traditional linear ARIMA model.BP neural network model embodies the advantages of artificial intelligence,its prediction accuracy is significantly higher than the traditional ARIMA model.And the BP neural network modeling method is more simpler than that of ARIMA model.The BP neural network need not set up complicated mathematical model,also do not need to understand the mathematical structure of the model and the correlation between the variables.Its applicability is very strong.ARIMA model is more suitable for relatively shortterm forecast analysis,as the prediction time,its forecast effect is declining gradually.AIDS incidence forecast in our country,adopting nonlinear BP neural network model is more appropriate.In the future in the study of gradually improve multi-dimensional input of the BP neural network,can make the model fitting and forecast more accurate,that will be have a big progress space.