When you make a forecast using Neural Network technology, you need always remember about some effect that takes place there. I could not find any definitions of this effect among available Neural Net technology information, so let us call this effect as "a reaction".
This is what it looks like:
Here we made an attempt to predict the close (not a relative price oscillator index). As you see, the projection line shows a sharp down right after a position of Learning Border Cursor.
This happens because we try to predict the indicator with some trend. However Neural Network technology is oriented to forecasting non trend indicators, like variations of some sinus wave. So this sharp down can be considered as a reaction of Neural Network on trending data. Neural Network prefers wave functions, so before making any forecast we need to convert our price history into something more suitable for the Neural Net.
Usually we recommend to use Relative Price Oscillator for this purpose. More about it you can read here: http://www.timingsolution.com/TS/FAQ/rpo.doc
Also I would like to mention that this is a standard math procedure, usually called "normalization"; and we have only one choice here: to find the best compromise between the oscillator and the real price.
Thus, when you see the sharp up/down movement right after the Learning Border Cursor, my recommendation to you is to be very cautious. Do this:
a) try to vary the Learning Border Cursor (LBC) position to be sure that this is just a reaction effect, not a forecast. In case of the reaction, the sharp drop (in our case) will move together with LBC.
b) try to use the relative price oscillator with a shorter period.