The last image isn't curve fitting -- it's showing the tradeoff of the true positive rate vs false positive rate as you vary the threshold of a binary classifier.