Juan PabloSoftware developerIf you can imagine it, we can build it.

COVID-19 ML Platform

A university machine learning project that predicts daily COVID-19 cases by country from about 330,000 records: a Random Forest model, a FastAPI service, a React interface and the whole system containerized with Docker Compose.

covid-19 ml platformCOVID-19 ML Platform: the site's home page
Status
Completed
Type
Academic
Since
May 2026
My role
Solo project: data preparation, model training, API, interface and containers.
Stack
Pythonscikit-learnpandasFastAPISQLAlchemySQLiteReactViteDocker

Problems and how I solved them

  1. 01

    ProblemA trained model adds no value if it only works inside an analysis notebook.

    SolutionI turned it into a service: the model is trained once, saved as an artifact, and a FastAPI service loads it to serve predictions to any application.

  2. 02

    ProblemPredictions and model performance had to be recorded so they could be reviewed later.

    SolutionEach prediction is stored in SQLite with its inputs, and the interface shows the history and the model's metrics (MAE, RMSE and R²).

  3. 03

    ProblemThe project had to run the same way on any computer, without installing dependencies by hand.

    SolutionBackend, frontend and training run in separate containers with Docker Compose; training runs as its own profile.

  4. 04

    ProblemThe initial evaluation reported an R² of 0.996: too good to be true.

    SolutionA closer review revealed two data leaks: the model received the same day's cumulative total, and the random split mixed neighboring days between training and test sets. With a date-based split and only past data, the real R² is close to 0.79.