Predictions a teacher can act on, with the reasons attached

What it does

Takes a student's learning history and predicts whether they are likely to struggle with an upcoming lesson, early enough to do something about it.

The pipeline

Feature engineering, three models compared against each other, and SHAP explainability so a prediction can be justified rather than just emitted.

That last part matters: a model that flags a student without being able to say why is not something a teacher can act on responsibly.

Served behind a documented REST API.