1

I am trying to understand what should I do when a models looses performance. Right now, when a model looses performance I am just creating a new model with new data but I’ve heard about the concept of calibrating a model. What does this mean? Does calibrating tackles the performance lost? How is it done?

Specifically I am talking about classification models (random forest, gradient boosting)

Thanks

mafalda
  • 11
  • 2
  • Calibration involves predicting the correct probability of class membership, essentially trying to get close to “pr” in my answer [here](https://stats.stackexchange.com/a/554551/247274). – Dave Dec 03 '21 at 22:06
  • Yes I always related calibration with score calibration, but I heard the term “calibration” in relation to deal with model performance lost and for my understanding it has something to do with parameters calibration (?) I just really would like to understand what can I do when a model in production looses performance – mafalda Dec 03 '21 at 22:14

0 Answers0