A customer churn predictor
The problem
Predict which customers of a subscription service will leave, using a few features, and decide whether the model is good enough to act on.
Requirements
- A reproducible training and test split
- A baseline and a k-nearest neighbours classifier on standardised features, scaled with training statistics only
- A confusion matrix, precision, recall and F1 on the test set
- A threshold or k chosen from the costs of missing a leaver and of a false alarm
Technologies
- Python 3
- Lists and dictionaries
- k-nearest neighbours
- Classification metrics
Step by step
- Profile the data and define the target and features.
- Split the data and standardise the features without leakage.
- Build the baseline and the classifier, and evaluate both on the test set.
- Choose the setting from the costs and write a recommendation.
Expected result
A short report recommending whether and how to use the model, backed by test-set results.
How it is evaluated
- No test information reaches the model
- The model is compared with a baseline
- Precision and recall are reported and explained
- The recommendation follows from the costs
Ways to improve it
- Try several values of k
- Add a feature and measure whether it helps
- Check the results for different customer groups