Prediction Of Epidemic Trend Of Hand,Foot And Mouth Pathogens Based On Time Series Analysis

Hand,foot and mouth disease(HFMD)is an infectious disease caused by different viruses,and most of patients are babies and children under 5 years old.The disease is highly contagious,and have complicated transmission routes,high epidemic intensity,and rapid transmission rate.Therefore,the prevention and control work is extremely difficult.Since 2010,HFMD has been ranked first in the incidence of legal infectious diseases in China.There are more than 20 enteroviruses that cause HFMD.The main pathogen of HFMD in China are EV-71 and CV-A16.In this paper,considering the transmission characteristics of HFMD,a time series long-term prediction model was constructed to provide scientific support for the formulation of prevention strategies for HFMD in advance.In this study,the monthly reported cases of HFMD in Shandong province and Guangdong Province from 2008 to 2016 were used as training data to construct a model to predict the incidence of HFMD in the months of 2017-2018 in the two provinces.RMSE(Root Mean Square Error)and MAE(Mean Absolute Error)were used to evaluate the prediction effect of the model.First,Seasonal Autoregressive Integrated Moving Average(SARIMA),a traditional time series model,was used for fitting and prediction.It is found that the SARIMA has a good prediction effect on data with strong regularity,but it has a poor prediction effect on data with a change in the propagation mode.Based on this,a combination of time series decomposition and SVR(Support Vector Regression)is proposed,the combined model first decomposes the hand-foot-andmouth disease time series to obtain long-term trend items,periodic seasonal items,and residual items,then use the SVR model to deal with nonlinear part(trends and remaining items),using four-fold cross-validation to select the hyperparameters of the SVR model,and then using the model obtained by fitting to predict the future trend items and residual items.The results show that the combination model is far more effective than SARIMA model in predicting the data when the propagation mode changes.In this study,the infectious disease dynamics SIR(Susceptible-Infectious-Recovered)model was further studied,due to the characteristics of seasonal transmission of hand,foot and mouth disease,the infectivity is set to a cosine form,combined with the HMM(Hidden Markov Models)model to estimate unknown parameters.First set the initial range of the unknown parameters,use Latin hypercube sampling to get 100 parameter sets,and bring the 100 groups of parameters into the model respectively.Use particle filtering to calculate the log-likelihood value,and then the optimal parameters which maximizes the logarithmic likelihood of the model is obtained by iterative filtering,using the optimal parameters to predict the future trend of hand-foot-mouth disease.The results show that the prediction effect of the SIR model is better than the combined model,and it can capture the pattern changes during the transmission of hand-foot-mouth disease.Since the dynamic model is effective in the long-term prediction of HFMD,we further apply it to the prediction of COVID-19.Considering the transmission characteristics of COVID-19,a SEIR(SusceptibleExposed-Infectious-Recovered)model was established to predict COVID-19 cases in wuhan.Compared to the SIR model,the SEIR takes exposed into account and quantifies population mobility and public interventions into the model.The number of cases reported from January 17 to February 11,2020 was used as the training data to predict the incidence of COVID-19 in wuhan in the following 29 days.The dynamic model of COVID-19 performs good in fitting and predicting,and the predicted peak incidence number and trend are consistent with the real data.The experimental results indicate that the basic reproduction number in the initial stage of COVID-19 is relatively high,and it will gradually decline to less than 1,then the epidemic will be gradually controlled,which is conform to the facts.