SVM comparing different chance levels

Software Support
,
1

Hi all,

I have three different experimental conditions, with different numbers of classes (2,3 and 4) and hence different chance levels (50%,33.33%,25%). What would be the correct approach to make these chance levels comparable, assuming that an above-chance accuracy scale with the chance level?

Thanks so much for any help in advance!

2

Hi Vivien,

Since oftentimes multiclass classification is based on multiple binary classifiers anyway, I would instead run all pairwise comparisons in the other classifiers and then average results. Else a direct comparison is just non-trivial.

Hope this helps!
Martin

3

Hi Martin,

Thanks so much, this helps!

All the best,
Vivien