A novel approach based on combining deep learning models with statistical methods for COVID-19 time series...

Area of Science:

• Epidemiology
• Data Science
• Machine Learning

Background:

• The COVID-19 pandemic necessitates accurate forecasting of infected cases for effective public health interventions.
• Deep learning models show promise for time series forecasting but can benefit from improved data representation.

Purpose of the Study:

• To enhance the performance of deep learning models for COVID-19 time series forecasting.
• To investigate the efficacy of time series augmentation techniques in improving forecasting accuracy.

Main Methods:

• Applied time series augmentation to generate synthetic data reflecting original series characteristics.
• Utilized augmented data to train and evaluate three deep learning models: Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Convolutional Neural Networks (CNN).
• Assessed model performance using Symmetric Mean Absolute Percentage Error (SMAPE) and Root Mean Square Error (RMSE).

Main Results:

• The proposed time series augmentation significantly improved the performance of LSTM and CNN models for COVID-19 forecasting.
• Average performance improvement was observed for GRU models.
• The study identified top-performing augmentation models and provided visual comparisons of actual versus forecasted data.

Conclusions:

• Time series augmentation is a valuable technique for enhancing deep learning-based COVID-19 case forecasting.
• The findings support the use of augmented data to improve the reliability of epidemiological predictions.
• The study offers insights into optimizing deep learning models for pandemic response.