Business Case

• A European bank with account holders in France, Spain, and Germany has had many customers churn recently and would like to decrease this churn rate.

Goals

• Understand what is driving customers and how this can be predicted in the data gathered on the customers before they churn.
• Create a predictive model to find customers who might churn so that special offers could be made to them to entice them to stay.

Task

• Train a decision tree binary classifier to create a predictive model of customer churn.

Insights

• The decision tree achieved the metrics of Accuracy 80.1%, sensitivity 83.1%, and specificity 68.6%.
• According to the decision tree, the top two variables that predicted churn were customer age and number of products with the bank.
• Indeed, 48% of the customers in the 45-54 age bracket had churned in this dataset. Also, 82% of the customers that had 3 products were in the churning group.

Recommendations

• The decision tree approach shows promise for prediction of churning customers.
• Customers that are predicted to churn could be targeted with offers to entice them to stay.

Technical Procedures

• Imbalanced classes in the training data were a problem (80%/20% split between non-churners and churners)
• To combat the problems caused by imbalanced classes, the minority class (churners) were oversampled with a SMOTE algorithm.
• The decision tree was trained using using the tidymodels packages in R.