|
Should I use PCA before put data in SVM? I think SVM can be trained better according to this http://en.wikipedia.org/wiki/Curse_of_dimensionality also maybe at all it will be faster? time(pca)+time(svm low dim)<time(svm high dim)? I also found this paper |
|
This is difficult to answer in general, since it depends on what kind of data you're dealing with. It may be a good idea in some cases, but if your data does not follow the assumptions necessary for PCA to make sense you might get worse results than feeding in the original data. There is a discussion of these issues in this tutorial: http://www.snl.salk.edu/~shlens/pca.pdf. If these assumptions make sense in your case, PCA may be a good idea, especially if you use additional unlabeled data to compute the PCs. |
|
If the test Oscar mentioned fails, you may also try to reduce the dimensionality of your data. This will effectively boost the performance characteristics of SVM. For instance, the use of self organizing maps (SOMs) are pretty common when combined with SVMs. Best, |
