Comparative COVID-19 Modelling
Data preparation, polynomial regression and a Lotka-Volterra model on five countries, evaluated with R², RMSE and MAE, and read with their limits.
- Role
- Student project, INP-HB
- Period
- Mar 2025
- Status
- Academic
- Context
- Senegal, Côte d'Ivoire, Burkina Faso, France, USA · Jan 2020 to Dec 2021
Engineering figures
- 3,550observations analysed
- 3 to 8polynomial degrees compared per country
- 1.01 to 2.52 %case fatality rates across countries
01Pipeline and processing
- Cleaning and correction of inconsistencies with pandas.
- Polynomial regression of cumulative cases (degrees 3 to 8, scikit-learn), best degree selected per country on R².
- A predator-prey Lotka-Volterra model adapted to infection dynamics.
- Evaluation with R², RMSE and MAE.
02Technology stack
- Python
- pandas
- scikit-learn
- Matplotlib
- Seaborn
- Plotly
- Jupyter
03Results and deliverables
- More gradual dynamics in the African countries studied than the sharper waves observed in France and the USA.
- Case fatality rates compared across countries.
04Limitations
- Exploratory curve fitting on cumulative cases, not a validated forecasting model.
- High polynomial degrees fit the observed period well but overfit and extrapolate poorly; a high R² on cumulative series is expected and says little about predictive power.
- Reported cases depend on each country's testing capacity, which limits cross-country comparison.