Variable selection using svm based criteria
Alain Rakotomamonjy · 2003
We propose some new methods for evaluating variable subset relevance in a purpose of variable selection . Relevance criteria are derived from Support Vector Machines and are based on the sensitivity of the weight vector kwk or the upper bounds of generalization error with respect to a variable. Experiments on linear and non-linear toy problem were ran and real-world datasets have also been used to assess the eectiveness of these criteria.