Confidence

The highest scoring state from the outputs from the neural network was taken, and the score of the next highest predicted state subtracted:

$$Confidence = integer (10 \times (out_{max}-out_{next}))$$ (1)

This is similar to the method exploited by PHD [17].