Use a metric to determine if the test data has been seen in training

[MedadRufus]

Use euclidean distance between test long/lat points and closest training data, to give an indication of closeness.
Perhaps, instead of merging by current, merge by long/lat, so that there are fewer datapoints