Dear Lalitha Kumari

1- If you don't have the required Matlab toolbox, you may use free svm toolboxes available online. I would recommend libsvm at: "http://www.csie.ntu.edu.tw/~cjlin/libsvm/".

Maybe you want to implement SVM, yourself. I would not recommend that. It could be a very complicated optimization problem and may do not converge well. Anyway you can take a look at:

http://www.cs.cmu.edu/~ggordon/SVMs/svm.m

and 

http://www.princeton.edu/~kung/ele571/571-MatLab/571svm/

2- COVER's theorem: http://en.wikipedia.org/wiki/Cover's_theorem

It simply says that if you map the data samples (which can not be separated by linear borders, i.e. they are not linearly separable) into another space with higher dimensions, the data samples with different labels "will" be separable by linear discriminating borders (hyper-planes) in the "probability sense" (i.e. when new dimensions go to infinity, the data samples with different labels are linearly separable with probability of 100%).

This mapping is not done explicitly. It is done implicitly (i.e. not really) using Kernel trick.

Find more at:

http://research.microsoft.com/pubs/67119/svmtutorial.pdf

and

http://www.support-vector-machines.org/SVM_osh.html

Dear Lalitha Kumari

1- If you don't have the required Matlab toolbox, you may use free svm toolboxes available online. I would recommend libsvm at: "http://www.csie.ntu.edu.tw/~cjlin/libsvm/".

Maybe you want to implement SVM, yourself. I would not recommend that. It could be a very complicated optimization problem and may do not converge well. Anyway you can take a look at:

http://www.cs.cmu.edu/~ggordon/SVMs/svm.m

and 

http://www.princeton.edu/~kung/ele571/571-MatLab/571svm/

2- COVER's theorem: http://en.wikipedia.org/wiki/Cover's_theorem

It simply says that if you map the data samples (which can not be separated by linear borders, i.e. they are not linearly separable) into another space with higher dimensions, the data samples with different labels "will" be separable by linear discriminating borders (hyper-planes) in the "probability sense" (i.e. when new dimensions go to infinity, the data samples with different labels are linearly separable with probability of 100%).

This mapping is not done explicitly. It is done implicitly (i.e. not really) using Kernel trick.

Find more at:

http://research.microsoft.com/pubs/67119/svmtutorial.pdf

and

http://www.support-vector-machines.org/SVM_osh.html

Fardin -- thanks for an excellent answer. I appreciate your reference to a good tutorial paper. It is helpful.

Thankyou Fardin sir.

Thanks for your suggestion Ahmadi sir

clc; clear; close all; %% Load Data xall=xlsread('D:\paper-tree\svm\in.xlsx'); yall=xlsread('D:\paper-tree\svm\out.xlsx'); xtr=xlsread('D:\paper-tree\svm\intr.xlsx'); ytr=xlsread('D:\paper-tree\svm\outtr.xlsx'); xtst=xlsread('D:\paper-tree\svm\intst.xlsx'); ytst=xlsread('D:\paper-tree\svm\outtst.xlsx'); TrainInputs=xtr; TrainTargets=ytr; n=numel(TrainTargets); %% Design SVM C=200; svmstruct=svmtrain(TrainInputs,TrainTargets,... 'boxconstraint',C,... 'kernel_function','rbf',... 'rbf_sigma',24); nTrainData=480; TrainOutputs=svmclassify(svmstruct,TrainInputs); plot (TrainOutputs,TrainTargets); mseTrain = mse(TrainOutputs,TrainTargets)

Hi

I need an SVM program (matlab code) for prediction.

thank you