Notes from a "Practical guide to SVM Classification"

A Practical Guide to SVM Classification

Recommended steps:

1. Transform data into SVM format
2. Scale data [either [-1,+1] or [0,1]
   
3. Try the RBF kernel function
4. Use cross-validation to find the best parameters - C (penalty parameter of the error term) and Gamma - a parameter for the RBF kernel function
5. Train with these parameters
6. Test

Issues:

1. Do not have enough data to perform cross validation to determine the parameters
2. Examples provided in paper show trivial datasets - 10's of features and 1000's of data items