Problem on forex prediction using neural netwoork

cicciospice

Hi all,
I would use some historical data on Forex (in my example I'm using EURUSD data) in order to predict future price oscillations.

What I have is 4 different prices: Open, Close, HIgh, Low per hour, starting from december 2012 until today, so a very large amount of data.

What I tryed to do is starting from 30 samples, in order to predict following 6 samples, so I  used windowing and I setted up operator with a window of "30" and an orizion of 6.

After several hours, Rapidminer crashes and I don't see any result.

I have a CSV file, correctly imported, with label on "date" column, and after windowing I see Close-30...Close-0, High-30...High-0 and so on, on y spreadsheet. Then I use a neural network but I don't change any parameter on it, so my forst question is: how many layers I should have?  ???

Maybe is better to have 4 different Neural Networks, instead of a big one with so many inputs?

Should I try to reduce input and try a dufferent window size (15, or 10)?

Any suggestion is very appreciated!
Francesco