Dual attention-based sequential auto-encoder for Covid-19 outbreak forecasting: A case study in Vietnam.

Area of Science:

• Epidemiology
• Data Science
• Artificial Intelligence

Background:

• COVID-19 prevention strategies include quarantine, medical treatments, and vaccination.
• Data-driven forecasting of COVID-19 cases is crucial for healthcare resource optimization and policy implementation.
• Recurrent Neural Network (RNN) models show promise but struggle with chaotic, non-smooth time-series data common in COVID-19 case counts.

Purpose of the Study:

• To propose a novel deep learning architecture for accurate COVID-19 case prediction.
• To address the limitations of existing RNN models in handling chaotic time-series data.
• To enhance temporal feature learning for improved outbreak forecasting.

Main Methods:

• Development of a dual attention-based sequential auto-encoding architecture (DAttAE).
• Integration of a dual self-attention mechanism within a Bi-LSTM based auto-encoder.
• Evaluation of DAttAE against traditional and state-of-the-art deep learning models using real-world datasets.

Main Results:

• The proposed DAttAE model demonstrates superior performance in learning and predicting new COVID-19 cases from chaotic time-series data.
• The dual self-attention mechanism effectively focuses on specific time-range sequences for enhanced prediction.
• Experimental results show DAttAE outperforms existing RNN-based architectures for COVID-19 outbreak prediction.

Conclusions:

• The DAttAE model offers an effective solution for predicting COVID-19 outbreaks using chaotic time-series data.
• Attention-based deep neural approaches provide significant improvements over traditional RNNs for epidemiological forecasting.
• Accurate forecasting supports better public health decision-making and resource allocation during pandemics.