time series prediction using neural network encog library java

Question

I read the example of Workbench Time-Series Example. But I am a little confused. I think in order to use the neural network I should have both the training and the evaluation dataset but what if I would like to predict future value that I don't have? For example suppose I have dataset that contains ten numbers and I make the first five numbers is the training data set and the rest is the evaluation data set and I need to predict the five numbers in the future after the ten number how I can do this? I hope this is clear thanks in advance.

Answer 1

Most probably ten numbers is not enough for making generalization. Training data should be representative in order to make good generalization. It can be compared to situation, if you want to make some conclusions with weather situation on planet while taking into account temperature only in one city during three months. You need to enlarge time period to at least five years and take into account much more cities around the world.