Solved – SVM low accuracy but significant p value

svm

I'm also new at doing SVM…

We have a data set of about 50 (25 in each group). We are trying to identify babies at a high risk of developing a disorder from babies that are low risk for developing a disorder. When I run SVM (LOOCV, linear kernal, ttest filtering), I'm getting really low accuracy levels (i.e. 24% in one case), but with statistically significant p-values. Can anyone help me understand why this is happening/what this could mean?

Thanks!

Best Answer

It's possible that you don't have enough data to reliably use an SVM. That, and, in my experience, they tend to perform best with high-dimensional data (e.g., 1000s of variables). How many predictor variable are you working with?

As a more direct answer to your p-value question, you should look at what the p-value is actually associated with--are you saying that it's associated with the accuracy statistic, or something else?

Related Question