I'm not an expert, but I imagine they would do at least these three things:
1. Correlate waveforms recorded by multiple phones in multiple places, i.e. look for a single fingerprint in the waveforms. If it's a bunch of different waveforms then it's probably not an earthquake.
2. For those waveforms that match the fingerprint: plot the velocity that the waves would have to be moving at in order to cause the waveforms in those particular places at those particular times and see if the speeds are consistent with the speed that earthquakes travel at and if the velocity vectors diverge from an epicentre. If the speeds are too low or too high or if the vectors do not diverge from an epicentre then it's not an earthquake,
3. Compare the waveforms with known waveforms from historical earthquakes, i.e. see if it looks anything like an earthquake.
In order to fool the system you would have to have multiple people in multiple places doing the same "silly walk" at just the right times.