Customer churn model
Predicting who will leave a bank is a ranking problem more than a prediction problem. This project treats it that way, and is honest about how much the model can say.
- 37%churn rate in the 1K customer records
- 52.6%of feature importance sits in age
- 0.60ROC-AUC, honest and modest
The question
Which customers are most at risk of churning, and how much can a model genuinely tell us?
What I did
- Prepared the customer data with BigQuery SQL.
- Trained a Random Forest classifier to estimate churn risk.
- Built Tableau views to show risk across customer groups.
- Packaged the results as LinkedIn carousel assets.
What I found
- ROC-AUC landed between 0.576 and 0.601 across runs.
- Age alone carried 52.6% of feature importance, and age plus income carried about 74% of the signal.
- That is only a little better than chance, so the model is useful for ranking risk, not for confident calls on a single customer.
Why I show it anyway
I would rather show a modest result clearly than dress it up. The sensible next step is better features and a cost-based view of which customers are worth contacting. Completed through the Lloyds Banking Group Data Science virtual experience on Forage. Not affiliated with or endorsed by Lloyds Banking Group.
- Random Forest
- BigQuery SQL
- Tableau
- Churn analysis