MS5 Fraud detection


I think we cannot come up with some simple algorithms to check email addresses or names or delivery addresses. Even if we can come up with some patterns, scammers can easily change their email or name. The delivery addresses are real so scammers can receive the good. (Maybe other users place orders in scammers website, and the scammers use hacked credit cards to buy things and send to real users with real addresses).
In this problem, I think we should use third party service. They have detailed data about ip address and email.
Because of the cost, I think we should apply this check when user sign-up, not when they place orders.

Sample data

I collected ip and email to check fraud score. You can checkout this link for the result.
I cannot collect more data because:
I get ip address from log, it’s only available for 1 month
only allows 200 API call for free plan

From this sample we can see:
Almost 99% of fraud users can be detected
Concern: 50% of normal users can be affected. (It’s really big but I think it will reduce much more if we check old orders(some orders in current list will be disputed) or a bigger sample. In non-dispute users, there are still a lot suspicious users)

Third party API

Currently, I am using to check fraud score of users’ email and ip address. I have not look in to other providers yet. It will cost us about 1000$ per month to check 500k new users.

Next steps

If we decide to use Third party API, there is next steps:
Collect more data from the past to see how many real orders get affected (If many normal orders be tagged as suspicious, it will be a huge burden for the CS team?)
Explore more similar services to pick the right one
2021-05-03 update:
I cannot run more test to verify but it seems the false-positive rate is quite high. There are many other providers, actually we cannot try them all. There is a list of those here with description and reviews
For a safe choice, I think we should try AWS fraud detector.

Want to print your doc?
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
) instead.