Forecasting COVID-19 new cases using NBEATS deep learning and mobility data

Area of Science:

• Epidemiology
• Computational Biology
• Machine Learning

Background:

• COVID-19 transmission is linked to human mobility.
• Accurate forecasting of COVID-19 cases is crucial for public health interventions.
• Existing models may have limitations in handling complex time-series data and long-term predictions.

Purpose of the Study:

• To propose and evaluate a novel deep learning approach, N-BEATS, for forecasting COVID-19 cases.
• To assess the impact of incorporating population mobility data on forecasting accuracy and interpretability.
• To compare the performance of N-BEATS against a state-of-the-art benchmark model (LSTM-Markov).

Main Methods:

• Utilized the N-BEATS deep learning architecture for time-series forecasting.
• Incorporated Google and Apple mobility data as covariates.
• Compared N-BEATS with LSTM-Markov using COVID-19 datasets from the US, UK, Russia, and Brazil.
• Evaluated model performance using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE).

Main Results:

• N-BEATS consistently outperformed LSTM-Markov across all tested datasets and countries.
• The N-BEATS model achieved lower RMSE and MAPE values compared to the benchmark.
• Incorporating mobility data as covariates significantly improved the accuracy and interpretability of the N-BEATS model.
• Mobility data were shown to provide substantial value for forecasting new COVID-19 cases.

Conclusions:

• The N-BEATS architecture is effective in capturing pandemic dynamics for COVID-19 case forecasting.
• Population mobility data are valuable covariates for enhancing the accuracy of epidemiological forecasts.
• The findings offer insights for policymakers and public health officials in managing infectious disease outbreaks.