I have a query on feature scaling for SVM. I understand the advantage and need to do feature scaling (gradient descent would be faster, less exposed to extreme values)
I am currently working on SVM and have feature scaled by the mean and sigma of each training data column (each feature). See Standardization
I am wondering is this the best means of feature scaling for SVM. Some SVM tutorial website suggests to scale to [-1, 1] instead? Would there be a difference. Trying to improve my classifier accuracy.