Project 21 of 39 · Data
HomeCredit Data Alchemy
Credit risk modelling on the Home Credit dataset, covering feature engineering across bureau and previous-application history through to model evaluation.
01Overview
A data preparation pipeline for the Home Credit default risk problem, turning seven raw relational tables into six domain-specific datasets built for modelling.
Most of the work is in the transformation: relational history becomes behaviour a model can actually use.
02How it is built
- Feature engineering driven by what the data means for credit risk rather than by what is easy to compute.
- Memory footprint reduced substantially without discarding information.
- Behavioural signals derived from mobile usage and application history.