Improved LSTM-based deep learning model for COVID-19 prediction using optimized approach

Area of Science:

• Epidemiology and Public Health
• Computer Science and Machine Learning
• Data Science and Predictive Analytics

Background:

• Infectious disease epidemics, such as the COVID-19 pandemic, cause significant global economic and physical disruption.
• Accurate forecasting of disease spread is crucial for effective public health interventions and resource allocation.
• Machine learning models are increasingly utilized for improved prediction of epidemic trends.

Purpose of the Study:

• To evaluate the performance of various time series forecasting models for predicting COVID-19 confirmed cases, deaths, and recoveries.
• To compare the efficacy of Long Short-Term Memory (LSTM) based models against other machine learning approaches.
• To assess the predictive capabilities of LSTM, Bidirectional LSTM (Bi-LSTM), and Gated Recurrent Unit (GRU) models.

Main Methods:

• Time series prediction using LSTM, Bi-LSTM, GRU, and dense-LSTM models.
• Evaluation of models on confirmed cases, deaths, and recoveries data from 12 major COVID-19 affected countries.
• Performance metrics included Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Median Absolute Error (MEDAE), and R2 score.
• Implementation using Tensorflow 1.0.

Main Results:

• LSTM-based models demonstrated strong performance in time series prediction of COVID-19 data.
• Comparative analysis showed LSTM models to be highly effective compared to other machine learning models evaluated.
• The models successfully forecasted annual trends, aiding in potential impact assessments.

Conclusions:

• LSTM-based models are highly effective and among the most advanced for time series forecasting of infectious disease data.
• The study validates the utility of deep learning approaches for predicting key COVID-19 metrics.
• Forecasting models are essential tools for understanding and managing the impact of epidemics.