Conclusion on p. 54 of Thoughtful Machine Learning (2015)

[imrek]

Hello,

I need some hint on how you arrived at the following conclusion:

"At this point, you'll notice we reduced the orders needing fraud review from 1,000 to 40 (i.e., 4% of the total).

Many thanks.

[hexgnu]

Hey there!

I believe you found an important issue which I'll throw into the errata.

The number of orders is 1000. 10% of those are fraudulent. 10% of those used a gift card. 60% of gift cards are fraudulent.

That means that out of the 100 fraudulent orders there are 60 that were matched using only the gift card. In the remaining 900 there are only 40 gift cards remaining. There are only 100 orders we need to look at not 150.

On the second paragraph we have reduced the need of review to only 40. Basically we only need to review 40 gift card orders but realistically we'd be reviewing all 100 anyways to double check.

Hope that helps thanks for submitting the issue.