<ul><li>The main purpose of calibration is to ensure that the model’s predicted probabilities should be consistent in real-life events.</li><li>Platt Scaling is a technique that is used to map logits to probabilities by using the sigmoid function.</li><li>Platt Scaling works well for binary classes, but when dealing with multi-class problems, One-vs-Rest(OvR) is applied to squash the multi-class problem to a binary class problem.</li><li>The CalibratedClassifierCV method is used for Platt scaling, which helps provide calibrated probabilities for each class, making predictions more reliable and confident.</li></ul>

Platt Scaling use in Multiclass Problems

Discover more