TheoreticalEcology / machinelearning

Teaching Material for a Machine Learning course with Keras and Tensorflow in R
https://theoreticalecology.github.io/machinelearning/
Other
17 stars 5 forks source link

Submission server - AUC #27

Closed Meier-Matthias closed 2 years ago

Meier-Matthias commented 3 years ago

We should force the users to upload 0/1 data at binary classification. The calculation of the AUC minds probabilities as well, but this is unfair compared to the ones who upload their classifications and not the probabilities. Example (the same data): AUC (0/1): 0.8629906637590297 AUC (probabilities): 0.9677618577075099

This makes no sense, if a definite classification is needed and not only a probability. Maybe we convert probabilities like (x<.5?0:1).