Using deep ensemble for influenza-like illness consultation rate prediction

The cases of influenza have always been an important global issue. Not only does it affect people's health, but also a crucial point in the policy implementation of government and health care units. If the trend of the epidemic can be predicted in advance, the public can be reminded to prevent it in advance to achieve control. This study aims to provide reliable forecasting services for changes in influenza-like epidemic trends. By collecting data on the emergency department visit rate of influenza-like illness, and linking the air pollution indicators and open data on temperature and humidity environmental factors, using the stacking method in the Ensemble Learning concept, first constructs deep learning models such like RNN, LSTM for the first stage training, and obtains the input data of the second stage model through cross-validation. This method combines the advantages of each model so that the model can have better performance. In addition, the outbreak calculation method published by the WHO is added to calculate the outbreak threshold of the influenza-like epidemic. This threshold will be used to determine whether the control status of the influenza-like illness is still in a reasonable control stage. The experiment confirmed that the model has a good prediction effect on the trend of the influenza like epidemic, and the evaluation index MAPE value can get an excellent performance of 8% to 15%. Finally, historical data and future forecast data are integrated on the web page for visual presentation to show the actual status of regional air quality and influenza-like data, and predict whether there is a trend of the influenza-like outbreak in the region. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

Influenza cases are a significant global issue impacting public health and policy implementation. Reliable forecasting models play a crucial role in predicting influenza-like epidemic trends, and the integration of Ensemble Learning and deep learning models can improve prediction performance. Evaluations show that the model has a good prediction effect, with an excellent MAPE value of 8% to 15%.