Data engineering for ML
Churn Prediction ML Pipeline
1
hard gate: failed validation stops the run before training
A small project about the principle that a model should never train on data you have not checked.
What I did
- Separated each stage into its own module. The validator checks required columns, nulls and numeric TotalCharges, and returns errors that halt the pipeline.
- Trains a random forest on one-hot encoded features and saves processed data and the model artifact.
Stack
Pythonscikit-learnPandas