Home

Welcome to the lecture_appml wiki!

Material that we want to look at to see what we can learn from them:

• https://twitter.com/predict_addict/status/1656922576763990021

  In my book 📕 “Practical Guide to Applied Conformal Prediction: Learn and apply the best uncertainty frameworks to your industry applications” -> https://a.co/d/0GhYNbz I am covering a lot of important business focused use cases including imbalanced data.

  One of the pitfalls of the most popular solutions such as over and under sampling, SMOTE is that they significantly harm calibration of the prediction models producing miscalibrated results.

  An interesting paper “The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression” reveals little known facts about imbalanced classed methods producing miscalibrated models.

  ”Results: The use of random undersampling, random oversampling, or SMOTE yielded poorly calibrated mod- els: the probability to belong to the minority class was strongly overestimated. These methods did not result in higher areas under the ROC curve when compared with models developed without correction for class imbal- ance. Although imbalance correction improved the balance between sensitivity and specificity, similar results were obtained by shifting the probability threshold instead.”