Balaji, Your question is confusing : you say that you are trying to classify the data with SVM. Then you ask about the kernel for support vector regression..

To answer your question, AFAIK, there is no automated methodology to select a kernel. Basically Gaussian (or radial basis) kernel should work for most data (continuous data). That said, if your data columns are probability distributions (probability simplex); then a kernel based on KL-Divergence is appropriate. If your data is nominal (or discrete) then you may want to consider a kernel base on Manhattan distance; if it consists of strings then of course string kernel... You can come up with your own kernel to suit your needs as long as it satisfies Mercer conditions (positive semi-definite)

Thank you for response.

Actually, I also wanted people to help me in Support Vector Regression

Hi...

You have to try different kernel functions. Basically it depends upon distribution of your data. So i suggest you to use WEKA data mining tool and change the kernel functions available in the tool. Then conclude.

Thank you Vipul sure I will look into it

For regression, in theory any kernel can perform well the task. However depending on the type of rule that you want to approximate, some kernel can perform more efficiently than other the task. It may even be the case where the kernel need not have a closed form as in: Eur. Phys. J. Special Topics. Vol. 143, 261-264 (2007).

Thank you Pedro can you please pop the link down here

Thank you Andre.

u try this out:

e1071 package of R.

Thank you Sajal

You have to examine different kernel functions and find the best one which produce less error.

The best way is to considering kernel function as a parameters , then optimized it with other hyper parameters such as C or Gamma. In the following example the RBF Linear and Poly Kernel are compared to each other.

https://scikit-learn.org/stable/auto_examples/svm/plot_svm_regression.html

https://scikit-learn.org/stable/auto_examples/model_selection/plot_grid_search_digits.html