Building a home roulette table and would like to forecast outcomes using pattern recognition

What kind of input / targets would patternnet need to forecast an output?

Just testing the theory here so keeping it simple and ignoring that each number has multiple stop points (like wheel of fortune), how can i go about trying to predict the outcome?

[Walter Roberson]

Are the stop points equal width or (like Wheel of Fortune) not equal width?

What kind of inputs are you thinking of for the forecast? Streaming video as the ball rolls around the outside?

Without something dynamic like that, it would not seem to be a problem suitable for pattern matching, just pure statistics, each outcome being (area of that outcome) / (total area of all outcomes). Unless, that is, you are supposing that there might be a correlation between runs and want the NN to figure out the correlation by feeding it a sequence of "real" outcomes.