Network structure and probability of prediction

I need to get a network with true/false (or 1/0, or 1/-1) output, which will be reacting with true only to the certain patterns of double data, and false to other "non-familiar" paterns of double data.
1. Help me please- what type of network, and what structure of layers should I use.
2. Is there something like Google's voice recognition "probability" of predicted result in Encog? I mean, can I get probability of result, given by Encog? and if I can- how exactly?