The Comparative Study Of Many Models On The Forecasting Incidence Of Syphilis

Syphilis have more transmission and infectious. Syphilis has great harm onlyafter AIDS in sexually transmitted disease, and could lead to AIDS. Syphilis couldcause great harm to the human and society, and the trend of syphilis is younger andaging. The syphilis has become a serious public health problem. The annual incidenceof syphilis ranked3after viral hepatitis and tuberculosis in notifiable diseases, and themortality rate of syphilis ranked7in notifiable diseases. The incidence of syphilis hasbeen increasing year from2001to2012, and the incidence rate of syphilis was30.44/100000in2012according《China Health Statistics Yearbook2013》. Forecatingincidence of syphilis might provide a scientific basis for the rational distribution ofhealth resourcesObjective:To arrange the epidemic characteristics of syphilis in China from2008-2013, andto explore the application of BP neural network, ARIMA, combined model, and theimproved time series decomposition model in the forecasting the monthly incidenceof syphilis in China of2013. We will select the best model through comparing thepredicted values and actual values of2013. We will use the best model to forecast themonthly incidence of syphilis in China of2014, and evaluate the syphilis epidemic of2014, which could provide a scientific basis for the rational distribution of healthresources.Methods:To get the monthly incident cases of syphilis and population from2008-2013through Chinese Center for Disease Control and Prevention and National Bureau ofStatistics of China, which could get the monthly incidence of syphilis, and the nuitwas1/100000. First, the incidence of syphilis of2008-2012was analysis through BP neural network, ARIMA, and combined model, then forecast the incidence tendencyof syphilis in China of2013. To improve the time series decomposition model, andforecast the incidence tendency of syphilis in China of2013. We will compare thefour forecasting models through SSE, MAE, MSE, MAPE, and the model is the bestwhich index is the least. Select the best model to forecast the incidence tendency ofsyphilis in China of2014. To judge the epidemic of syphilis in China of2014throughthe forecasting of overall trend with reference range and the maximum and minimumtrend.Results:(1) BP neural network model network structure is as follows: in the past threeyears, the incidence of the same period as the network input data, earlier onset of thecurrent rate of the network output, and the hide layer is10. Compare the forecastvalue and actual value of2013, SSE=1.4362, MAE=0.2418, MSE=0.0999,MAPE=10.20%.(2) In ARIMA, p=1, d=1, q=1,P=0, D=1, Q=1, S=12, and theparameters of ARIMA(1,1,1)(0,1,1)12has statistically significant. Compare theforecast value and actual value of2013, SSE=1.1299, MAE=0.2404, MSE=0.0886,MAPE=9.97%.(3) The weight1=0.41889of ARIMA and the weight2=0.58111of BP neural network in combined model. Compare the forecast value and actualvalue of2013, SSE=0.9998, MAE=0.1784, MSE=0.0833, MAPE=7.79%.(4) We usedthe improved time series decomposition model (multiplication process) to forecastmonthly incidence of syphilis in China of2013. Compare the forecast value andactual value of2013, SSE=0.3851, MAE=0.1497, MSE=0.0517, MAPE=5.97%.(5)The seasonal adjustment factor of improved time series decomposition model(multiplication process) was0.99831. Compare the forecast value and actual value of2013, SSE=0.3735, MAE=0.1456, MSE=0.0509, MAPE=5.82%.(6) We used theimproved time series decomposition model (addition process) to forecast monthlyincidence of syphilis in China of2013. Compare the forecast value and actual value of2013, SSE=0.3752, MAE=0.1330, MSE=0.0510, MAPE=5.55%.(7) The seasonaladjustment factor of improved time series decomposition model (addition process) was0.06042. Compare the forecast value and actual value of2013, SSE=0.3634,MAE=0.1302, MSE=0.0502, MAPE=5.44%.(8) The forecasted monthly incidence ofsyphilis in China of2014was1633/100000,2.1047/100000,2.8185/100000,2.6471/100000,2.7542/100000,2.7253/100000,2.8357/100000,2.7699/100000,2.6090/100000,2.3858/100000,2.4161/100000,2.3877/100000used the improvedtime series decomposition model (multiplication process).(9) The forecasted monthlyincidence of syphilis in China of2014was2.2243/100000,2.1494/100000,2.8056/100000,2.6383/100000,2.7514/100000,2.7072/100000,2.8210/100000,2.7624/100000,2.5953/100000,2.3853/100000,2.4220/100000,2.3829/100000usedthe improved time series decomposition model (addition process).Conclusion:(1) The prediction accuracy of forecasting monthly incidence of syphilis usedimproved time series decomposition model was better than BP neural network,ARIMA and combined model.(2) The prediction accuracies of forecasting monthlyincidence of syphilis used improved time series decomposition model(multiplication process and addition process) was better after adjusting seasonal factor.(3) Application of the improved time series decomposition model to predict theincidence of syphilis in2014, compared with the actual values, and have higherprecision.(4) The prediction accuracies of forecasting monthly incidence of syphilisused multiplication process and addition process were all better, and the prediction ofaddition process was better slightly.(5) The forecasted epidemic of syphilis of2014was in the control based on the forecasting of overall trend with the reference rangetrend.