How to perform multi-class classification using 'svm' of e1071 package in R
Solution 1
The iris dataset contains three class labels: "Iris setosa", "Iris virginica" and "Iris versicolor". To employ a balanced one-against-one classification strategy with svm, you could train three binary classifiers:
The first classifier's training set only contains the "Iris setosa" and "Iris virginica" instances. The second classifier's training set only contains the "Iris setosa" and the "Iris versicolor" instances. The third classifier's training set--I guess by now you'll know already--contains only the "Iris virginica" and the "Iris versicolor" instances.
To classify an unknown instance, you apply all three classifiers. A simple voting strategy could then select the most frequently assigned class label, a more sophisticated may also consider the svm confidence scores for each assigned class label.
Edit (This principle works out of the box with svm
):
# install.packages( 'e1071' )
library( 'e1071' )
data( iris )
model <- svm( iris$Species~., iris )
res <- predict( model, newdata=iris )
Solution 2
R document says that "For multiclass-classification with k levels, k>2, libsvm uses the ‘one-against-one’-approach, in which k(k-1)/2 binary classifiers are trained; the appropriate class is found by a voting scheme."
Comments
-
StrikeR over 1 year
I want to perform multi-class classification using the
svm
function ofe1071
package. But from what I came to know from the documentation ofsvm
, it can only perform binary classification. The vignettes document tells this for multi-class classification: "To allow for multi-class classification,libsvm
uses the one-against-one technique by fitting all binary subclassifiers and finding the correct class by a voting mechanism".
What I still don't understand is if we can perform the multi-class classification withsvm
ofe1071
in R? If yes, please explain how we can do it overiris
dataset. -
StrikeR about 10 yearsthat is a very nice way of approaching the problem. But, before going that far, I want to know if the
svm
ofe1071
can directly perform this multi-class classification. -
StrikeR about 10 yearsthe reason I'm asking is that, if I have a data with 10 output classes I need to model 10C2 (10 combination 2) = 45 such classifiers, which is a huge task.
-
idleherb about 10 yearsOk :) So the short answer is yes,
svm
can also do multiclass classification and it works the same way as for binary (see edit in my answer). -
Eric Hauenstein about 6 yearsQuoting the documentation doesn't really answer the question.