(a) Distribution of neural network predictions on the training set. (b) Confusion matrix when a threshold is set at t = 0.5. Quadrants in black represent the number of correct predictions for each class and the white quadrants the number of mistakes for each class.