Research On ARIMA-SVR In Prediction Of Tuberculosis Incidence

Pulmonary tuberculosis is a chronic pulmonary infectious disease caused by Mycobacterium tuberculosis,and droplet transmission is the main way of transmission.The most fundamental and effective measure of modern TB control strategy is to effectively improve the detection rate and cure rate of TB patients.It is of great practical significance to accurately predict the incidence rate of tuberculosis,formulate reasonable target rate and high-risk season.The development trend of pulmonary tuberculosis shows a combination of linear and nonlinear characteristics.Considering the advantages of support vector machine regression model（SVR）in processing small sample data and the characteristics of differential autoregressive moving average（ARIMA）in fitting periodic data,this thesis uses ARIMA prediction model,SVR prediction model,ARIMA-SVRseries model andARIMA-SVRparallel prediction model to match the monthly incidence trend of tuberculosis in China.Using the data of tuberculosis incidence from 2009 to2018 to fit the trend of tuberculosis incidence,the optimal model was selected,and the number of tuberculosis monthly incidence from 2019 to 2020 was predicted,providing reference basis and model suggestions for the national tuberculosis prevention and control work.Firstly,this thesis describes the research background and significance of tuberculosis,the research status at home and abroad,and the prediction ability of common infectious diseases prediction model.At the same time,it describes the theoretical basis and modeling steps of linear prediction model and support vector machine nonlinear prediction model in detail.Secondly,this thesis takes the monthly incidence of tuberculosis in China from2009 to 2017 as the training set,and the monthly incidence of tuberculosis in China in2018 as the test set to establish a prediction model to predict the incidence trend of tuberculosis from 2019 to 2020.Time series linear prediction model（ARIMA）,support vector machine nonlinear prediction model（SVR）,ARIMA-SVRseries combination model andARIMA-SVRparallel combination model are used to fit the national tuberculosis incidence trend.Genetic algorithm（GA）and particle swarm optimization（PSO）are used to optimize the parameters to select the best parameters.Three comprehensive indexes（RMSE、MAE、MAPE）are used to describe the comprehensive prediction accuracy of the model,and absolute error（AE）and relative error（RE）are used to describe the prediction effect of specific sample data.The comprehensive error index of the modelARIMA（1,1,1）（2,0,0）₁₂is MAE（28）4492.1 16,MAPE（28）.5000%,RMSE（28）5748.5 83.The comprehensive error index of the SVR model is MAE（28）3531.279,MAPE（28）3.82%,RMSE（28）5409.772.The comprehensive error index ofARIMA-SVRparallel model is MAE（28）2758.36 4,MAPE（28）3.10%,RMSE（28）4452.775.The comprehensive error index ofARIMA-SVRseries model is MAE（28）4492.1 15,MAPE（28）.4996%,RMSE（28）5748.30 6.TheARIMA-SVRparallel combination model effectively combines the advantages of linear prediction of time series and nonlinear prediction of support vector machine,and improves the prediction accuracy by 38%and 18.42%on the basis of the model.It can be seen that the combined model is better than the linear model and the nonlinear model in predicting the trend of monthly incidence of tuberculosis in and the parallel combination modelARIMA-SVRand parameter optimization method（PSO）have a slight advantage in predicting the incidence trend of tuberculosis in China.Finally,the thesis summarizes the current epidemic trend,the prediction results and feasibility of tuberculosis,points out that the incidence rate of pulmonary tuberculosis is slow and the awareness rate of public TB knowledge is low.It proposes the suggestions for increasing prevention and control in spring season and usingARIMA-SVRparallel model in medium and short term prediction.