I am working on a multi-label learning classification problem. When I tune the hyperparameters of the model, the sign of each label (each line on the MLL) remains the same while the scores change.
For example, an instance is predicted 1 in the first test with 0.5 score for label one and 0.4 for label two. While on the second, it's also predicted as 1 but with a less score for label 1 than the second.
As I am interested on the value of the score, I would like to know if there is a method that can tell me which of the two tests is better. In other words, I would like to say that this label is more likely to be predicted than the other (even though that a test gives better score to other one).