Stable predictions for health related anticausal prediction tasks affected by selection biases: the need to deconfound the test set features

no code implementations • 9 Nov 2020 • Elias Chaibub Neto, Phil Snyder, Solveig K Sieberts, Larsson Omberg

In health related machine learning applications, the training data often corresponds to a non-representative sample from the target populations where the learners will be deployed.