Ensemble Learning with Evolutionary Computation: Application to Feature Ranking

Abstract. Exploiting the diversity of hypotheses produced by evolutionary learning, a new ensemble approach for Feature Selection is presented, aggregating the feature rankings extracted from the hypotheses. A statistical model is devised to enable the direct evaluation of the approach; comparative experimental results show its good behavior on non-linear concepts when the features outnumber the examples.

• Type: Conference
• Year: 2004
• Where: PPSN
• Authors: Kees Jong, Elena Marchiori, Michèle Sebag
• Keywords: Distributed And Parallel Computing, Feature Rankings, Feature Selection, Non-linear Concepts, PPSN 2004