It's common knowledge in DS that overfitted models perform well on training data and poorly on test data. But how do you decide if a model is really overfitting? I have nowhere (books, online courses, internet literature) found a prescribed threshold for defining a model as overfitted.
Is there a threshold for test data? Say in a classification model accuracy (or relevant evaluation metric like recall or F1 score or RMSE in regression) for test data should be at least 75% or 80% or some percentage X of training data. So if training data has accuracy of say 90% then model is not overfit as long as test data has accuracy of at least X% of 90%. So is there any such X threshold prescribed.
Looking forward for an answer from Data Scientists in the community. Thanks