A Deep Learning Framework for Using Search Engine Data to Predict Influenza-Like Illness and Distinguish Epidemic and...

Area of Science:

• Public Health
• Epidemiology
• Data Science

Background:

• Seasonal influenza epidemics pose a significant global health threat.
• Web search data offers valuable insights for forecasting epidemics.
• Existing prediction models often overlook the nuanced sensitivity of search terms to seasonal changes.

Purpose of the Study:

• To propose a deep learning framework for predicting influenza epidemic states.
• To leverage Baidu search index and influenza-like illness (ILI%) data for enhanced forecasting.
• To differentiate prediction accuracy across various epidemic states.

Main Methods:

• Collected weekly ILI% data (2013-2024) and corresponding Baidu search indexes.
• Performed cross-correlation analysis between search queries and ILI%.
• Developed and evaluated a Convolutional Long Short-Term Memory (CLSTM) network for 1-3 week ahead predictions, comparing performance across different time periods (all, epidemic, nonepidemic).

Main Results:

• ILI% exhibits regular seasonal incidence in China.
• Predictions were more accurate when dividing data into epidemic and nonepidemic seasons (MAPE 10.73%) compared to the all-time period (MAPE 12.78%).
• The CLSTM model integrating Baidu search data outperformed models using ILI% alone and demonstrated superior performance over LSTM and transformer models in specific scenarios.

Conclusions:

• Combining Baidu index with traditional surveillance data enhances influenza prediction accuracy, especially when considering distinct epidemic and nonepidemic seasons.
• The developed CLSTM framework offers a promising approach for improving influenza early warning systems.
• This research provides a novel perspective for public health preparedness and timely disease response.