COVID-19 in Iran: Forecasting Pandemic Using Deep Learning.

Area of Science:

• Epidemiology
• Computational Biology
• Data Science

Background:

• The COVID-19 pandemic rapidly spread globally, necessitating accurate forecasting methods.
• Predicting disease outbreaks is crucial for public health resource allocation and intervention strategies.

Purpose of the Study:

• To forecast the COVID-19 outbreak trajectory in nine specific countries using deep learning models.
• To evaluate and compare the performance of various deep learning models for pandemic forecasting.

Main Methods:

• Applied deep learning models: multilayer perceptron, random forest, and long short-term memory (LSTM) variants.
• Utilized COVID-19 occurrence data, country information (names, population, area) for model training.
• Collected data from January 22 to July 30, 2020, for training and August 1-31, 2020, for testing.

Main Results:

• A modified LSTM model (M-LSTM) demonstrated superior performance in forecasting.
• The M-LSTM model achieved high accuracy with metrics: MAPE (0.509), RMSE (458.12), NRMSE (0.001624), and R² (0.99997).
• Investigated the impact of country-specific actions on model predictions by halting training at key dates.

Conclusions:

• The M-LSTM model provides accurate and reliable COVID-19 outbreak predictions.
• Deep learning approaches, particularly M-LSTM, are effective tools for epidemiological forecasting.
• Understanding the influence of public health interventions on forecast accuracy warrants further investigation.