Chapter 17 Margin Based Feature Selection and Infogain with Standard Classifiers
Ran Gilad-Bachrach, Amir Navot · 2006
The decision to devote a week or two to playing with the feature selection chal- lenge (FSC) turned into a major effort that took up most of our time a few months. In most cases we used standard algorithms, with obvious modifica- tions for the balanced error measure. Surprisingly enough, the na¨ ive methods we used turned out to be among the best submissions to the FSC. Using the insights we gained during our participation in the challenge we developed novel feature selection algorithms (G-flip and Simba), which are based on the large margin principle . We used early versions of these algorithms during the FSC, a detailed description of the mature version of G-flip and Simba can be found in our contribution to part III of this book (Chapter 29).