Between-subject (patients versus controls) binary SVM classifier with unequal sample sizes

1

Dear TDT team,

I want to run a binary SVM (or other classifier) discriminating patients versus controls based on their first-level con images, with unequal sample size (n=96 versus 65).

Is this scenario implemented in TDT and if yes, could you please help me setting up this analysis?

Thanks very much in advance!

Best wishes,

Lukas Van Oudenhove

2

Hi Lukas,

Yes, TDT supports this! Check out the templates folder and look at the between group template. Then mix that together with the unbalanced data template. I think unless you want to do subsampling, I would simply use AUC_minus_chance as a metric instead of accuracy_minus_chance and use the between group decoding. If you run into any errors, I hope that TDT will guide you but it should work out of the box mostly!

Hope this helps, and happy decoding!
Martin