Deep Learning-Based Prediction of Daily COVID-19 Cases Using X (Twitter) Data

Area of Science:

• Computational epidemiology
• Social media analytics
• Machine learning for public health

Background:

• COVID-19 control necessitates innovative predictive methods.
• Social network data offers a promising avenue for real-time disease surveillance.
• Existing epidemiological models can be enhanced by incorporating diverse data streams.

Purpose of the Study:

• To predict confirmed COVID-19 cases using Twitter data and deep learning.
• To evaluate the efficacy of a Time Series Mixer (TSMixer) model for multivariate time series prediction.
• To assess the potential of social media data for epidemiological forecasting.

Main Methods:

• Data extraction from X (Twitter) using natural language processing (NLP).
• Utilizing daily COVID-19 G-value (growth rate) from worldometer as the target variable.
• Development and evaluation of a Time Series Mixer (TSMixer) deep learning model.

Main Results:

• Achieved a Mean Squared Error (MSE) of 0.0063 for 24-month G-value prediction.
• Optimized prediction using MinMax normalization, Recursive Feature Elimination (RFE), and aggregation methods.
• Demonstrated the model's effectiveness on a multivariate time series dataset.

Conclusions:

• Social media data, specifically tweets, can significantly enhance daily COVID-19 case predictions.
• The TSMixer model shows potential for accurate epidemiological forecasting.
• Integrating social network data offers a valuable tool for public health surveillance and disease control.