Data Scientist project
Build a churn prediction model and evaluate it correctly
A churn model on a real public dataset — graded on evaluation choices under class imbalance, not just on training a model.
The brief
Build a churn prediction model using a real public dataset (for example, the Telco Customer Churn dataset on Kaggle, or a comparable public employee-attrition dataset). Do exploratory data analysis, engineer at least a few features beyond the raw columns, train a baseline model and a stronger model, and evaluate both using metrics appropriate for an imbalanced classification problem (not accuracy alone).
Suggested stack
What you hand in
- A public GitHub repository with the notebook(s) and a README
- The README states the dataset source and how to reproduce the results
- A short section stating the final model's precision, recall, and ROC-AUC, and what trade-off was chosen for the classification threshold
Grading happens against the rubric below, so read it before you start — not after.
How this is graded
Published in advance and weighted out of 100. Nothing here is a surprise.
The EDA surfaces real patterns in the data (e.g. which features correlate with churn) and at least some engineered features are genuinely derived, not just the raw columns renamed.
The imbalance is identified explicitly and addressed (resampling, class weights, or threshold tuning), with the reasoning stated.
Precision, recall, and ROC-AUC (or equivalent) are reported and interpreted correctly for this imbalanced problem — accuracy alone is not treated as sufficient.
A reader can run the notebook top to bottom and get the same result; cells are ordered logically with markdown explaining each step.
No console errors or crashes, no broken layout, no leftover placeholder text or commented-out code.
Why this project is worth your weekend
- Churn is a canonical real business problem with skewed classes, exactly where junior candidates default to reporting accuracy and get it wrong.
- Using a well-known public dataset lets an interviewer sanity-check the reported numbers instantly against known benchmarks.
- The EDA-to-feature-to-model narrative is what a real take-home actually evaluates, not just the final model's score.
Where people lose points
- Reporting accuracy on an imbalanced dataset where predicting "no churn" for everyone already scores 80%+.
- Leaking future information into a feature (using a field only known after the churn event actually happened).
- No train/test split discipline — evaluating the model on data it was trained on and reporting an inflated score.
Other Data Scientist projects
Two or three of these turn an empty resume into a portfolio.
Built it? Get it scored against this rubric.
Submit your work and get a score on every criterion above, written feedback, and three resume bullets you can use straight away.