Prediction of COVID-19 Using a WOA-BILSTM Model

Area of Science:

• Epidemiology
• Computational Biology
• Artificial Intelligence

Background:

• The COVID-19 pandemic necessitates accurate infection number prediction.
• SARS-CoV-2 transmission is influenced by temporal and spatial factors.
• Neural networks are increasingly used for epidemiological forecasting.

Purpose of the Study:

• To propose and evaluate a novel whale optimization algorithm-bidirectional long short-term memory (WOA-BILSTM) model for predicting cumulative confirmed COVID-19 cases.
• To compare the performance of the WOA-BILSTM model against various established machine learning and statistical models.

Main Methods:

• Inputting regional epidemic data including cumulative, cured, death, and daily cases.
• Utilizing a bidirectional long short-term memory (BILSTM) network as the base model.
• Incorporating the whale optimization algorithm (WOA) to optimize BILSTM parameters.

Main Results:

• The WOA-BILSTM model demonstrated superior accuracy in predicting cumulative confirmed COVID-19 cases compared to LSTM, BILSTM, GRU, CNN, CNN-LSTM, RNN-GRU, DES, ARIMA, linear, Lasso, and SVM models.
• Experiments conducted using data from Beijing, Guangdong, and Chongqing, China, confirmed the model's effectiveness in complex scenarios.
• The WOA optimization algorithm showed fast convergence and ease in finding optimal solutions compared to Bayesian and grid search algorithms.

Conclusions:

• The proposed WOA-BILSTM model offers a highly accurate and efficient approach for COVID-19 case prediction.
• The model's performance suggests its utility in assisting governments with developing effective public health control measures.
• The study highlights the potential of hybrid AI models, combining optimization algorithms with deep learning, for epidemiological forecasting.