Teaching Your Machine to Find Fraudsters

Very interesting talk on a topic not usually covered, i really enjoyed the linearity in teaching the different solutions, so going from the simplest and showing the downfalls and advantages and then progressing to other solutions, very well delivered.

Easily the best presentation of the conference. Clear, thorough, and useful; good pacing and diagrams. Absolutely excellent.

This is a very rich talk with a lot of potential.

Give a definition of Fraud up front, and the different types of fraud you're attempting to mitigate during your talk. Right now this is in the middle of your talk. Move all the fraud types to the front, then the progress through that list and show how to mitigate each one. This will give the audience a roadmap for what you're going to talk about, and if they're not familiar with the problem domain, it can help bring them in.

The click fraud stuff is awesome, and you could probably do a talk on that alone, but it's very dense with code and math. The code itself is hard to follow as you're talking about it. Some of this was due to code formatting and variable naming, but I understand you're trying to condense it to fit on the slide. This might be better abstracted across several slides as simple math equations, instead of code, and displayed in the same order as your code uses.

Some of my feedback may be due to my own unfamiliarity with the problem domain, so keep that in mind too :)

Great delivery!

Great talk, again clearly knowledgeable on the subject, good pacing and good delivery. I will definitely investigate the svm extension for my next classification need. Thank you.

Absolutely solid talk on detecting fraud. I liked that you went over some red flags for fraud that many people over look.

good talk, lots of examples - really makes you think