What is more significant?

Leave some Data out of training and run test against previous unseen data, or run test 1000x against DETREND = SHUFFLE?;