Exploring Feasibility of Multivariate Deep Learning Models in Predicting COVID-19 Epidemic

Area of Science:

• Epidemiology
• Computational Biology
• Data Science

Background:

• Mechanistic models for COVID-19 (coronavirus disease 2019) face challenges due to evolving understanding and data limitations.
• Accurate and comprehensive data are crucial for effective epidemiological modeling of complex pandemics.
• Rapidly refreshing knowledge of COVID-19 complicates traditional modeling approaches.

Purpose of the Study:

• To develop a data-driven workflow for modeling the COVID-19 epidemic using deep learning (DL).
• To create an alternative modeling approach that complements existing mechanistic models.
• To extract, process, and apply DL methods to COVID-19 time-series data.

Main Methods:

• Extensive data extraction and annotation from over 60 official press releases in Hubei, China (2020).
• Development of multivariate Long Short-Term Memory (LSTM) models for predicting COVID-19 time series (1, 2, and 3 days ahead).
• Comparison of multivariate LSTMs against univariate LSTMs for tracking key COVID-19 metrics.

Main Results:

• A comprehensive 10-variable dataset was processed for 125 days in Hubei.
• Multivariate LSTM demonstrated good predictability for new deaths, hospitalizations, discharges, and monitored patients.
• Multivariate LSTM outperformed univariate LSTM for 1-day-ahead predictions of new and total cases, and new deaths.

Conclusions:

• Deep learning models are feasible for complementing mechanistic approaches in infectious disease modeling.
• DL models provide a valuable alternative when epidemiological mechanisms are still under investigation.
• The study highlights the potential of data-driven workflows in pandemic modeling.