Bank customer churn

I used a dataset of 10,000 bank customers to find which characteristics were associated with churn, then trained a model to predict it. Inactive customers and customers aged 41–70 left more often; customers with two products left less often.

A random forest reached about 85% accuracy and 46% recall on customers who left. Recall is the number that matters here, because a model that always predicts "stays" is already about 80% accurate.