Support vector machines - what's the point?

rogojel

TS Contributor
#1
hi,
I just finished the chapter on SVMs from the Statistical Learning book of Hastie and Tibshirani. Their focus is to use the SVMs for classification - and they also show that the SVMs are equivalent to the logistic regression with nonlinear predictors (there is even an impressive exercise to drive the point home)

Given that many more people will be familiar with logistic regression and that the logistic regression handles the nonlinearities in a much more transparent fashion, why would anyone ever chose an SVM as a classification method?

regards
 

hlsmith

Not a robit
#2
I have had similar thought. I say but did not read an article on SVM regression the other day, I wish I could find it. Though, see the following two parts or three link for some insights. I always felt the interpretability and generalization of SVM was more limited.


http://www.datasciencecentral.com/p...c-regression-vs-decision-trees-vs-svm-part-ii


http://www.datasciencecentral.com/profiles/blogs/logistic-regression-vs-decision-trees-vs-svm-part-I


P.S., in about a month I am going to start reading that book from cover to cover, until then I am still drowning in work :(