There are two versions of one-class SVM. One algorithm is to find a close
boundary supported by SVs that cover most target samples, developped by D.
Tax; another is try to separate target samples from origin as well as
possible, developped by B. Schölkopf. You may find more information at
David Tax's homepage at http://www-ict.ewi.tudelft.nl/~davidt/
Be sure to read his excellent Ph.D thesis.
Hope it helps
Yun
|