Enhanced bat algorithm for COVID-19 short-term forecasting using optimized LSTM.

Area of Science:

• Computer Science
• Epidemiology
• Artificial Intelligence

Background:

• The COVID-19 pandemic necessitated accurate forecasting to manage its spread.
• Existing machine learning models for COVID-19 case prediction often lack optimized temporal and nonlinearity handling.
• Effective forecasting is crucial for pandemic control until vaccines are widely available.

Purpose of the Study:

• To develop an optimized long short-term memory network (LSTM) for enhanced COVID-19 case forecasting.
• To improve the accuracy and reduce the mean absolute error of COVID-19 predictions.
• To address limitations in temporal components and nonlinearity in current forecasting frameworks.

Main Methods:

• Proposed an optimized long short-term memory network (LSTM) model for COVID-19 case forecasting.
• Applied the bat algorithm (BA) for LSTM optimization, introducing an enhanced variant to overcome premature convergence and local minima.
• Incorporated Gaussian adaptive inertia weight and Gaussian walk for improved swarm intelligence and local search mechanisms within the BA.

Main Results:

• The optimized LSTM model demonstrated superior performance in forecasting COVID-19 cases.
• Achieved a high accuracy of 99.52% in experimental evaluations.
• Outperformed non-optimal LSTM, recurrent neural networks, gated recurrent units, and other state-of-the-art algorithms.

Conclusions:

• The proposed optimized LSTM, enhanced with a modified bat algorithm, offers a highly accurate method for COVID-19 case forecasting.
• This approach effectively addresses temporal dynamics and nonlinearity issues in pandemic prediction.
• The findings suggest a valuable tool for public health strategies in managing infectious disease outbreaks.