Class distribution reflected in AUC?

by Har23   Last Updated December 06, 2018 14:19 PM - source

I'm learning on churn models and i found a curious result. I tried a logistic regression and random forest model and used k-fold and ROC metric. I created a model for each group of customer (20). Each customer has a different class distribution. The churn % per group descends when we are in the next group. churn % in group 1 (40%) > churn % in group 2 > ...> churn % in group 20 (20%). The amount of the samples of Churn+NonChurn has the same trend : 1>2>...>20. When I plot the AUC per group I see that the AUC follows the same trend and descends when we have more imbalanced data. is it makes sense?

Tags : roc auc metric churn

Related Questions

Measuring customer attrition after a branch closure

Updated August 01, 2017 19:19 PM