Forecasting influenza activity using self-adaptive AI model and multi-source data in Chongqing, China.

Area of Science:

• Epidemiology
• Artificial Intelligence
• Public Health

Background:

• Timely influenza detection and response are crucial for public health preparedness.
• Existing studies often focus on regional/national levels with regular trends, lacking local-level forecasting for irregular patterns.
• Accurate local influenza forecasting remains a challenge.

Purpose of the Study:

• To develop and evaluate an innovative AI model for forecasting local influenza activity.
• To address the limitations of existing methods in predicting irregular influenza trends.
• To improve preparedness for seasonal influenza epidemics and pandemics at the local level.

Main Methods:

• A Self-adaptive AI Model (SAAIM) was developed, integrating Seasonal Autoregressive Integrated Moving Average (SARIMA) and XGBoost models.
• SAAIM utilized multi-source electronic data: historical influenza-like illness (ILI%) percentage, weather data, Baidu search index, and Sina Weibo data from Chongqing, China.
• The model's forecasting performance for ILI% in Chongqing (2017-2018) was compared against three existing models.

Main Results:

• ILI% in Chongqing exhibited irregular seasonal trends from 2012 to 2018.
• SAAIM demonstrated superior performance in forecasting ILI% compared to reference models, achieving a mean absolute percentage error (MAPE) of 11.9%, 7.5%, and 11.9% for 2014-2016, 2017, and 2018, respectively.
• Historical influenza activity data was the most significant contributor to forecast accuracy, followed by weather data and internet public sentiment data.

Conclusions:

• The Self-adaptive AI Model (SAAIM) enables accurate influenza forecasting in locations with irregular seasonal patterns.
• Multi-source electronic data integration is effective for enhancing local influenza prediction.
• This approach supports improved public health strategies for managing influenza outbreaks.