Ground Truth and training data set

Your training data is what you train your classifier on.

You then test the accuracy of your model on your test set.

Ground truth refers to the label for each training sample you have i.e. you know which category/outcome each training sample belongs to