I worked on this project when I was a Data Scientist at Datank.
Goal
- Predict which borrowers will make late payments for a Credit Risk team
Duration
- Six-month project
Activities
- Wrangled a very messy transaction stream
- Recognized and extracted relevant information required for feature engineering and visualization
- Made the selection of Machine Learning models by assessing their perfomance metrics
- Applied exponential decay to features, improving dramatically the precision of the predictions
- Refactored -cleaning, training and prediction- code into Dockerized tasks as required by the Data Engineering team
Toolbox
- Docker
- Python
Outcome
- Achieved model performance of 80% precision (99% accuracy)
- Coded appropriate solutions fit for production
- An API that delivers predictions seamlessly to a Credit Risk team
)