J.T. de Souza, N. Japkowicz, and S. Matwin (Canada)
Feature Selection, boosting, machine learning.
The problem of feature selection is defined as follows: given a set of candidate features, select a subset that per forms best under some classification system. A large number of feature selection systems have been developed in the last few years, most of them with very good results with regard to increasing classifier accuracy. This paper proposes a technique aimed at boosting the performance of feature selectors based on some element of randomness. The idea is to run the feature selector a number of times and compare the different lists of features produced by these runs, retaining (using a random but guided approach) as the most relevant ones the features that appear more frequently. The method is conceptually simple, easy to implement, and yields very encouraging results.
Important Links:
Go Back
