Back then I was basically just using a logistic regression node to predict all the price returns of EURUSD, USDJPY and NZDUSD based on all the indicators. So for example predict EURUSD on the indicators of the other two.
It’s very important that you set the thresholds right (does a forecasted return of e.g. 1% trigger a trade or not), as the predicted return must be large enough to cover the bid-ask spread.
I’ve now used to a slightly more complex setup, that I currently don’t want to share.