Round up/down a neural network output

I'm currently predicting match outcomes using neural network backpropagation, the target value should be either a 1 (win) or 0 (loss). My neural network output almost always has a value in between those two values, so is it possible to convert values that are below 0.5 into 0s and values that are above into 1s so it's easier to distinguish if it's a win or a loss?

Also, how to code the percentage accuracy of the network?

Thanks in advance.