Introduction to Pattern Recognition: A Matlab Approach

58 CHAPTER 2 Classifiers Based on Cost Function Optimization

almost equal values for the λ

i

’s. Thus, the summation in the classification rule in Eq. (2.14)

exhibits low variation for the various x’s (from both the training and the test set), which leads

to reduced discrimination capability. Values between these two ends lead to more acceptable

results, with the best performance being achieved for σ = 1 in the present example. The

previous discussion makes clear the importance of choosing, for each problem, the right

values for the involved parameters.

• Third, for fixed kernel parameters and C varying from 0.2 to 20,000, the number of SVs (in

general