Question regarding load the trained model and test on original dataset

Here is the case:
 
For a dataset, I used 10-fold cross-validation to train a Neural Network model, the EER I get is around 15%. The NN model is saved.
 
I wish to generate the predications using the saved model for the original whole dataset. The results is surprisingly high, EER only 1%.
 
I have thought the saved model could only give me a result close to 15%.
 
Could you please give any comments how to get an objective result using the saved model?
 
Thanks!
 
Henry

_______________________________________________
Wekalist mailing list
Send posts to: Wekalist@...
List info and subscription status: https://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html