Regularization for a svm classifier
17 views (last 30 days)
I am applying Regularization in SVM for classification, but I cannot find a specific command to do it. Is there a command for this? or I should apply SVM in one line then regularized it in the other line? I found "fitclinear" and I can apply regularization, but there is no way to add different Kernel to this command.
Vishal Bhutani on 10 Sep 2018
By my understanding you want to train a SVM for classification with regularization. As you might be aware that ‘fitclinear’ creates a linear classification model object that contains the results of fitting a binary support vector machine to the predictors X and class labels Y. And it allows to set regularization parameter. For different kernel like RBF or Gaussian, you can try ‘fitckernel’. Here is the documentation link for that:
Another thing you can try is using ‘fitcsvm’ and in that you can vary box-constraint parameter for regularization and also you can vary kernel functions. More regularization (small boxconstraint) means you allow more points in the margin. With no regularization (large box constraint, hard margin) you allow no data points in the margin. Regularization relaxes this constraint, and the smaller boxconstraint the wider the margin but the more data points lie within the margin. Here is the documentation link for that: