Research And Application Of Error Correction Model Based On Data Preprocessing In Respiratory Infectious Diseases Prediction Of China

Objective Since respiratory infections caused serious harm to human health,in order to avoid further expansion of their impact and damage to the quality of life,it is necessary to prevent the outbreak of respiratory infectious diseases.Therefore,based on the CEEMD method,this study combines parameter optimization,weight search and error correction methods to establish an error correction prediction model.And this study used the time series data of the monthly incidence of tuberculosis,scarlet fever and pertussis reported in China from 2004 to 2017 to verify the accuracy and reliability of the error correction prediction model.This study aims to provide a feasible prediction method for respiratory infectious diseases with different epidemic characteristics,and then provide predictive information for relevant departments to prevent and control the respiratory infectious diseases.Methods First,the original time series data（monthly incidence of tuberculosis,scarlet fever,and pertussis）were decomposed into different intrinsic mode functions（IMF）using the CEEMD methods.And removed IMFs from the original time series data to obtain the remaining component sequence.Second,PSO-SVR,GSA-SVR,PSOGSA-SVR,GRNN and LSSVM were used to fit the remaining components,and IMFs were fitted with PSO-SVR.After selecting the lag order（Lag）and the number of IMFs,added the prediction results of the IMFs and the remaining component sequence to obtain 45 individual error correction models.Finally,according to the model evaluation indicators,selected the optimal individual error correction model,and judge whether it met the requirements of the prediction effect.If the evaluation index of the optimal individual error correction model did not meet the requirements,then a combined error correction model was established.Results（1）The prediction accuracy and reliability of the optimal individual error correction model and combined error correction model based on CEEMD decomposition were higher than that of the parallel model.And the MAE,MSE,R and IA of the two error correction models were better than the parallel model,and the R and IA of the two error correction models were above 0.96.（2）The combined error correction model had better prediction effect than the individual error correction model,and the overall prediction trend of the combined error correction model among the three respiratory infectious diseases was closer to the true trend,especially extreme points.（3）The individual error correction model selected based on data-driven and model evaluation indicators had different numbers of IMF and Lag orders.In this study,the best fitting model of the remaining component of the three types of respiratory infectious diseases that constitute the individual error correction model was also different.Conclusions（1）The error correction model based on data preprocessing had significantly improved the prediction accuracy and reliability of respiratory infectious diseases incidence data,and the prediction effect of the combined error correction model was higher than that of the individual error correction model.（2）Three types of respiratory infectious disease incidence data with different epidemic characteristics were used to verify the accuracy and reliability of error correction model,the prediction results proved that the error correction model had high generalization performance and could be applied to predict respiratory infectious diseases with such time series characteristics.（3）The error correction model which integrated a variety of methods,such as data preprocessing,parameter optimization,weight search and error correction,was better than every single model and parallel model.