COVID-19 Case Prediction Model Comparison

This page compares weekly COVID-19 case predictions from four different models: LSTM/GRU (Long Short-Term Memory/Gated Recurrent Unit), ARIMA (AutoRegressive Integrated Moving Average), Random Forest, and XGBoost. The graphs below show our predictions against the actual reported cases and compare the performance of all models.

Model Performance

The performance of each model is measured using the Mean Absolute Percentage Error (MAPE). Lower values indicate better performance.

Final Seven Weeks: Predicted vs. Reported Cases

This chart displays the actual cases and predicted number of COVID-19 cases for the final seven weeks of reported data (weekly totals, July 7 to August 18, 2024), comparing all four models: LSTM/GRU, ARIMA, Random Forest, and XGBoost.

Methodology

We used four different models for time series forecasting:

1. LSTM/GRU (Long Short-Term Memory/Gated Recurrent Unit) neural network model
2. ARIMA (AutoRegressive Integrated Moving Average) model
3. Random Forest Regressor
4. XGBoost Regressor

All models were trained on COVID-19 case data from the WHO dataset. Our prediction pipeline followed these steps:

1. Daily data collection from the WHO COVID-19 dataset
2. Data preprocessing and cleaning
3. Feature creation with a sequence length of 90 days
4. Model training with the latest data
5. 7-day hindcast generation for all models
6. Comparison of predictions with actual reported cases
7. Calculation of Mean Absolute Percentage Error (MAPE) for each model

For more details on our methodology, please visit our GitHub repository.