Related Experiment Video
Updated: Aug 13, 2026
10:05
High-throughput Detection Method for Influenza Virus
Published on: February 4, 2012
Influenza-Like Illness Forecasting Using Multisource Data: Comparative Deep Learning Study
Caixia Dang1,2, Yeqing Tong3, Yanquan Mo1,2
1Chinese PLA Center for Disease Control and Prevention, South Gate, Yard 20, Dongda Street, Fengtai South Road, Fengtai District, Beijing, 100071, China, 86 18149338893.
JMIR Medical Informatics
|August 11, 2026
Summary
Integrating digital data like internet searches and mobility significantly improves influenza-like illness (ILI) forecasting. Long short-term memory (LSTM) models, incorporating these sources and policy stringency, offer superior predictive accuracy for public health surveillance.
Area of Science:
- Epidemiology
- Data Science
- Public Health
Background:
- Accurate forecasting of influenza-like illness (ILI) is vital for effective public health interventions.
- Traditional surveillance methods can be enhanced by integrating novel digital data streams, including internet searches and human mobility.
- Optimal modeling frameworks for multisource data integration in ILI forecasting remain an area for exploration.
Purpose of the Study:
- To develop and compare various data-driven models for forecasting ILI incidence trends using multisource data.
- To evaluate the predictive performance of different machine learning models, including SARIMA, LSTM, CNN-LSTM, Transformer, and random forest.
- To identify the most influential data sources and model configurations for accurate ILI forecasting.
Main Methods:
- Weekly ILI incidence data and multisource variables (temperature, humidity, air quality, Baidu Search Index, Baidu Migration Scale Index, Oxford Stringency Index) for Hubei Province (2020-2023) were collected.
- Models were trained on an 85:15 split, optimized using grid search with 5-fold cross-validation.
- Performance was evaluated using MAE, RMSE, MAPE, and R² metrics.
Main Results:
- Multisource models significantly outperformed univariate benchmarks, with LSTM achieving the highest R² (0.8350) by integrating environmental, mobility, and internet search data.
- The Oxford Stringency Index consistently reduced prediction errors across all models, decreasing MAPE by up to 84.87% in the LSTM model.
- The Baidu Search Index was the most influential single external predictor, while model performance varied with feature composition.
Conclusions:
- Deep learning models, especially LSTM, effectively utilize heterogeneous digital data for improved ILI forecasting.
- Internet search trends and policy stringency are critical external predictors for ILI prediction.
- The developed LSTM model demonstrated robust external validation and may be applicable to similar regions in central China.
Related Concept Videos
01:18Steps in Outbreak Investigation
In the ever-evolving field of public health, statistical analysis serves as a cornerstone for understanding and managing disease outbreaks. By leveraging various statistical tools, health professionals can predict potential outbreaks, analyze ongoing situations, and devise effective responses to mitigate impact. For that to happen, there are a few possible stages of the analysis:
01:27Influenza
Influenza is an acute, highly communicable viral disease that affects the respiratory tract and is responsible for seasonal epidemics worldwide. Influenza A is the most prevalent type associated with widespread outbreaks and is subtyped based on two surface glycoproteins: hemagglutinin (H) and neuraminidase (N), as in H1N1. These glycoproteins are essential for viral infectivity, transmission, and immune recognition. Transmission occurs primarily through respiratory droplets and contaminated...