We need to create a classifier that is able to achieve an acceptable level of accuracy. Possibly we should have a low number of false positives (so that users of our tool will only see differences that are more likely to be differences), so we should aim for a high precision.

Let's say we aim for 70% precision.