[PLenz]

Not really fraud, phones have different levels of horizontal accuracy based off of privacy settings, GPS & cell signal strength, battery life, CPU load and that's before you get to the obfuscation that exchanges do. At Dstillery we built a geodata classifier to tell us what was and what wasn't good data. We throw out 60% to 75% a day as not useful for learning anything from. But we can be picky since we combine web and location data we aren't beholden to needing the unreasonable amounts of location data you need working with just location data. Location should be holistic part of the data, not the be-all, end-all.

[PLenz]

It should also be noted that we have anti fraud tech baked into our system that fires before our geodata stuff runs. Fraud gets cleared out for being fraud not for being bad location data.