I am working on a problem that involves a mix of counts, proportions and dollars. The problem is in the e-commerce fraud space. Customer orders are placed and scored by a classifier to determine their probability of being fraudulent. Based on some manually set threshold score, the orders are either approved or declined, leading to an approval rate which can be measured in count (Approved Order Count / Total Order Count) or sum (Approved Order Sum (\$) / Total Order Sum (\$)).
Inevitably, some fraud gets through in the approved orders. We can measure this using the chargeback rate (Chargeback Sum (\$) / Approved Order Sum (\$)).
I am trying to model the chargeback rate ($) as a function of the approval rate (in count). I have collected the daily approval rate count and chargeback rate $.
What would be a good way to model this? I’m not sure how to deal with the mixture of counts and sums in this problem.