I am using a large number of features (around 200K) which are mostly zero with the occasional 1 and am performing a binary SVM classification task with this data. I was wondering what kind of kernels would be useful fo this data. Normalizing the data to have mean of zero and std of one is out of the question as it would stop the data being sparse.
I have tried all the usual kernels RBF, Linear etc ... and am happy with the results, I just want to know if there are any specific kernels for sparse binary features as there are for string etc ..
[link] [6 comments]