Next: Introduction Up: home page
In this tutorial we present a margin based feature selection criterion. This criterion is used to devise selection algorithms for multi-class categorization problems. We present the theoretical generalization bound for this setting. We present a comparison of the margin based feature selection algorithms to other feature selection schemes such as Relief and R2W2 and report promising experimental results.
This tutorial is based on the authors work presented in [GBNT04] and [GBNTar].