"Cross validation for One-Class SVM models"
Hi, I've just uploaded the process X-Validation with One-Class SVM to myExperiment. The basic idea is to cross validate One-Class SVM models by partitioning the data as usual (for instance, into 10 parts), to train the classifier only on the examples of one class, but to test on both classes (for the part that was left out for testing).
I'm very interested in feedback. For more stable validation results, one certainly should perform the whole X-Validation several times and take the average of the individual results. For simplicity, I left this step out. But do you see any other problems or errors with this setup, please?
Best wishes
Marco
I'm very interested in feedback. For more stable validation results, one certainly should perform the whole X-Validation several times and take the average of the individual results. For simplicity, I left this step out. But do you see any other problems or errors with this setup, please?
Best wishes
Marco
Find more posts tagged with
Sort by:
1 - 1 of
11
the only problem I can imagine is that the actual train size differs from fold to fold. This might bring in additional instability, so I would increase the number of outer iterations compared to the usual number.
Greetings,
Sebastian