Multi-step prediction for influenza outbreak by an adjusted long short-term memory

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Influenza causes significant global morbidity and mortality annually.
• Accurate forecasting is crucial for healthcare resource allocation and vaccine manufacturing.
• Existing models lack sufficient accuracy for multi-step-ahead influenza prediction.

Purpose of the Study:

• To develop and validate a multi-step-ahead time-series forecasting model for influenza outbreaks.
• To enhance the accuracy of predicting influenza-like illness (ILI) rates.
• To provide a flexible tool for hospitals and pharmaceutical companies.

Main Methods:

• Utilized four different multi-step prediction algorithms within a long short-term memory (LSTM) network.
• Implemented a six-layer LSTM structure with multiple single-output predictions.
• Evaluated model performance on US influenza-like illness rates for predictions ranging from two to 13 steps ahead.

Main Results:

• The six-layer LSTM with multiple single-output predictions demonstrated superior accuracy.
• Achieved mean absolute percentage errors (MAPE) below 15% for 2- to 13-step-ahead predictions.
• The average MAPE across all predictions was 12.930%.

Conclusions:

• The developed LSTM model offers a highly accurate method for multi-step influenza forecasting.
• This approach represents a novel application and refinement of LSTM for influenza outbreak prediction.
• The methodology holds potential for global application in influenza prevention and control efforts.