Can anybody explain how to perform one-class svm
clasification in WEKA.
I have weka 3-5-3 version.It's working good.
I'd recommend to upgrade to the latest CVS version of Weka, LibSVM
contains a few bugfixes there (as well as probability estimates):
My work is on FRAUD DETECTION and my data set contains
33 attributes out
of which one is fraud lable(fraud found or not found).
How to apply 1-class svm to that data set?
Since you have actual 2 classes (fraud and non-fraud), I'm not sure why
you wanna use the 1-class SVM (on the other hand, I'm not too familiar
with the libsvm terminology, so I might be wrong)?
Do I need to remove the fraud label from the data
I have divided the data set into two parts.
Part1 contains entire fraud data(ie fraud label is yes) and part2
contains entire non fraud data(ie fraud label is no).Two parts has 33
attributes and order of them is same.
Can anybody suggest me the solution.
My main problem is I dont know how to set the training and test data
into WEKA for 1-class svm classifications.
Why don't you just put all the data in one file and then let Weka do a
train/test split (10-fold cross-validation) on it with a C-SVC/nu-SVC
SVM? Then you wouldn't have any problems with non-matching headers (the
class attribute of the train set will only have "fraud" as label,
whereas the test set is gonna have "non-fraud", the way you're splitting
Peter Reutemann, Dept. of Computer Science, University of Waikato, NZ
+64 (7) 838-4466 Ext. 5174