Regional Influenza Prediction with Sampling Twitter Data and PDE Model

Area of Science:

• Epidemiology
• Computational modeling
• Public health surveillance

Background:

• Geotagged Twitter data offers opportunities for real-time flu epidemic monitoring.
• The sheer volume of social media data necessitates effective data sampling strategies.
• Predicting flu trends accurately from sampled social media data remains a challenge.

Purpose of the Study:

• To develop a real-time influenza prediction method applicable to sampled social media data.
• To create a partial differential equation (PDE) model that accounts for flu dynamics and interventions.
• To evaluate the model's predictive accuracy under various data sampling scenarios.

Main Methods:

• Simulated the data sampling process for flu-related tweets.
• Developed a mechanistic partial differential equation (PDE) model to characterize flu tweet volumes.
• Incorporated factors of flu spread, recovery, and public health interventions into the model.

Main Results:

• The PDE model demonstrated high prediction accuracy, exceeding 90% with only 1% of sampled Twitter data.
• Even with aggressive sampling (0.1% and 0.01%), the model achieved relative accuracies of 85% and 83%, respectively.
• The model effectively mitigated the impact of data reduction caused by sampling.

Conclusions:

• The developed PDE model is highly effective for predicting temporal-spatial flu trends from limited sampled Twitter data.
• This approach offers a robust solution for real-time public health surveillance of infectious diseases.
• The mechanistic model shows promise for overcoming challenges in social media data-driven epidemic prediction.