Thesisretention Keep Sample data

Step 1 of 5

How the numbers are defined

Active subscribers
People who have not canceled.
At risk
Active subscribers with a score of 0.45 or higher.
Precision
Of the highest-risk tenth of the holdout, the share who had canceled.
Recall
Of the holdout subscribers who canceled, the share with a score of 0.45 or higher.
Holdout
The last 30% of the sample. The model is not fit on these people. Precision and recall describe the score. They do not change the active-subscriber count.