Bank Marketing: Who to Call

problem · A bank sells term deposits by phone, and only about 12% of clients say yes. Calling everyone is expensive. Which clients are worth the call?

approach & results

We compared logistic regression, a decision tree, a random forest and gradient boosting in Python on 45,211 bank clients.
Gradient boosting ranked likely buyers best, with a test AUC of 0.9248, and its training and test scores were only 0.0028 apart, so it wasn't overfitting.
Recommended scoring every client and calling only the top 10%, about 4,521 people, which cuts around 90% of the calls.