Comparative Analysis Of Prediction Effect Of ARIMA,RNN And BIRNN Models On Influenza In Jiangxi Province

Objective:1.Based on Cite Space software,knowledge mapping analysis was used to summarize the influenza prediction methods,and to illustrate the progress,hotspots and future trends of various prediction methods in influenza prediction.2.An examination of the epidemiological features of influenza-like illness(ILI)occurrences in Jiangxi province,based on the surveillance data of ILI cases,is being conducted.3.Based on the data of influenza-like illness,three prediction models were established and compared by using R software and Python software,to supply Jiangxi province with a scientific theoretical foundation for influenza prevention and control.Method:1.English articles on population influenza prediction methods included in the core collection database of Web of Science from January 2003 to February 2023 were searched.Microsoft Excel 2016 and Cite Space 6.1.R6 were used to analyze the number of publications,countries and keywords.2.The monthly surveillance data of influenza-like illness in Jiangxi province from2004 to 2018 were collected and analyzed by descriptive statistical methods.3.From January 2004 to December 2017 the influenza data was used as the training set,while January 2018 to December 2018 was used as the test set.Subsequently,the monthly surveillance data of influenza-like illness in Jiangxi province from 2004 to 2017 was used to construct the SARIMA(Seasonal Autoregressive Integrated Moving Average)model.Py Charm of Python 3.9 software was then employed to construct the RNN(Recurrent Neural Networks)and BIRNN(Bi-directional Recurrent Neural Networks)models.Mean absolute error(MAE),root mean square error(RMSE)and mean absolute percentage error(MAPE)were used to evaluate the predictive ability of the three models.Result:1.A total of 351 articles were retrieved in the field of influenza prediction from2003 to 2023.The trend of publications: from 2003 to 2015 was a slow period,the number of publications increased from 0 to 24 per year;From 2016 to 2023,the number of papers published will increase to 51 per year at most.There were 25 emergent keywords related to influenza prediction methods.In the early stage,"impact","mortality" and "pneumonia" were the main keywords,and in the recent stage,"influenza-like illness","deep learning" and "machine learning" were the main emerging keywords,and "impact" ranked first.2.In Jiangxi province,from 2004 to 2018,an average annual incidence rate of1.72% was recorded for 87306 cases of influenza-like illness(ILI).There was an obvious epidemic trend of influenza in 2009 and 2018.The seasonal periodicity of influenza in Jiangxi Province is obvious,and the cycle is one year.The peak of influenza in Jiangxi Province is seen between December and March of the following year,with a secondary peak occurring in July.This is the primary season of high incidence,with winter and spring being the transition,and summer having a slight peak.3.Three prediction models were established based on the training set data:SARIMA(0,1,2)(0,0,1)12 model,RNN model,and BIRNN model.There was a slight deviation between the predicted trend of SARIMA model and the real number of influenza-like illness cases,which could be used as a reference for the future trend of influenza incidence in Jiangxi province.The trend of influenza predicted by RNN and BIRNN model was basically consistent with the trend of the real number of influenzalike illness,which could well predict the trend of influenza incidence in Jiangxi Province.In terms of prediction performance,the BIRNN model had the best prediction performance,with MAE,RMSE and MAPE of 97.2,117.29 and 7.54,respectively,which were reduced by 11.82%,33.05% and 82.28% compared with the SARIMA model,respectively,It showed significant improvement.Combining MAE,RMSE,MAPE and prediction graph,the prediction performance of RNN model and BIRNN model was better than that of SARIMA model,and BIRNN model had the best prediction performance.Conclusion:1.Through the analysis of knowledge graph method,it is found that the prediction methods for influenza population surveillance data have gradually transitioned from traditional regression models and time series models to machine learning methods with higher prediction performance,and developed from a single model to a multi-factor model.2.The SARIMA model’s prediction performance was significantly outdone by RNN and BIRNN models,particularly the latter,which had a higher accuracy and was more precise in forecasting the occurrence of influenza-like illness in Jiangxi province,thus forming a solid theoretical foundation for local influenza prevention and control.3.By comparing the three models,it was found that the prediction performance of the neural network model was better.It was hoped that this kind of prediction model could be applied to the establishment of the influenza prediction network in Jiangxi Province in the future,so as to provide more accurate theoretical guidance for the prevention and control of influenza in Jiangxi Province.