2.2 What it does

The way the FOIA predictor works is that it takes a big long list of FOIA requests - successful and not successful - and calculates a bunch of measurements for each. How many sentences are in it, how long an average sentence is, if you reference FOIA fees, and so on.

It then looks at each of those metrics - “feature,” we call them - to see how they’re related to whether a request is successful or not.

Then you take your own request and feed it to the predictor. The predictor calculates all the same metrics as it did on the other requests - sentence length, etc - and uses them to predict whether your request will be successful or not.

When the FOIA Predictor is used on requests it hasn’t seen before (but requests that we know whether they were successful or not), it’s right about 80% of the time, which sounds amazing.

It’s that 80% accuracy rate that we’re interested in now.

First, let’s see how that accuracy rate is calculated. We’re going to be following along with the FOIA Predictor’s code, because wonderfully enough, it’s all open source. We’ll talk more about that later.