Question about the ROC curve

[SilasK]

Hello,

I'm using Codacore and find it to be a very interesting tool.

You provide an easy function to plot the ROC curves of the ratios identified. These ROC curves are based on the entire dataset given to Codacore, which is the training dataset.
If I understand correctly, ROC curves and other metrics derived from the training dataset are not very meaningful, are they?
Several times, I have been disappointed by the poor performance on a holdout dataset.