Why use Platt's scaling?

In order to calibrate a confidence level to a probability in supervised learning (say to map the confidence from an SVM or a decision tree using oversampled data) one method is to use Platt's Scali...