Neural network accuracy for character recognition

Question

I have implemented a handwritten character recognition system in Matlab using neural networks. I am confused on how to calculate the accuracy of the system. Do I have to consider the performance on testset only? Or, do I need to calculate the overall accuracy using training set + test set?

Answer 1

The accuracy of the system is considered to be how the trained network performs on the test set, ie, a set it hasn't seen during training.

The accuracy on the training set is only used during training, to update the weights and potentially determine when to exit training.