CODEWITHSUNDEEP
matlabmachine-learningsvm
Ankita
Ankita

Reputation: 485

Prediction using SVM Regression?

I want use Support Vector Machine (SVM) for prediction. And with I have written code as follows using matlab function fitrsvm and predict,

tb = table(x,y)                                                  
Mdl = fitrsvm(tb,'y','KernelFunction','gaussian')                                                                          
YFit = predict(Mdl,tb);                                
scatter(x,y);                                                   
hold on                                                
plot(x,YFit,'r.')

The output I am getting here .
Here blude is testing values (tb) and red is prediction using SVM. As you can clearly see this prediction is wrong. Could anyone tell me any way to improve the prediction close to the measured values ?

Upvotes: 4

Views: 4065

Answers (1)

FesianXu
FesianXu

Reputation: 447

you should use Kernel Function like RBF or gaussian and so on. enter image description here

the default Kernel of the SVM is K(xi, xj) = xi*xj and it is a linear kernel.Of course you can only get a linear regression result.

Code like

x = 0:0.01:5 ;
y = sin(x)+rand(1, length(x)) ;
x = x' ;
y = y' ;
tb = table(x,y) ;
Mdl = fitrsvm(tb,'y','KernelFunction','gaussian');
YFit = predict(Mdl,tb);                                
scatter(x,y);                                                   
hold on                                                
plot(x,YFit,'r.')

=======================================================================
As for the accuracy of the result, it dependst on many factors like the type of Kernel, the punish coefficient adjustment or something else, it usually needs for times to adjust the parameters. cross-validation could help you to find a good set of parameters

Upvotes: 3

Related Questions