SVM Regression

Can you clarify why support vector machine regression is not included in MATLAB Tool box? However SVM Classification is included in Bioinformatics Toolbox. Is there any specific reason or disadvantage in using SVM for Regression? Advance Thanks.

Answer 1

[VladMel]

You can extent an output of svmclassify function by changing a line:

function outclass = svmclassify(svmStruct,sample, varargin)

to

function [outclass v]= svmclassify(svmStruct,sample, varargin)

and change a line(115):

classified = svmdecision(sample,svmStruct);

to

[classified v]= svmdecision(sample,svmStruct);

v - is regression value

[zohar]

+1

Answer 2

[Amith Kamath]

I don't really know why MATLAB has not included it by default, and as far as I know, there is no disadvantage as such of using SVM for regression, just that there are other better known methods. SVM classification is much more popular, especially in the bioinformatics field and hence I suppose they have decided to have a native implementation. Nevertheless, you will find a very good implementation of SVM r in this toolbox:

http://www.csie.ntu.edu.tw/~cjlin/libsvm/

[Royi Avital]

How do you use it for regression?

Answer 3

[Richard Willey]

The SVM implementation in Bioinformatics Toolbox does not support Support Vector Regression.

With this said and done, the boosted and bagged decision tree functions in Statistics Toolbox both support nonparametric regression and can be used to solve many of the same types of problems. Alternatively, if you have access to Neural Network Toolbox this is another obvious way to skin the cat...

I'd be very interested to know whether there are any specific problems where an SVM implementation generates better results than boosted or bagged decision trees.