Information-theoretic algorithm for feature selection
Pattern Recognition Letters
Feature subset selection using a new definition of classifiability
Pattern Recognition Letters
| Hi-index | 0.00 |
Feature selection serves for both reduction of the total amount of available data (removing of valueless data) and improvement of the whole behavior of a given induction algorithm (removing data that cause deterioration of the results). A method of proper selection of features for an inductive algorithm is discussed. The main idea consists in proper descending ordering of features according to a measure of new information contributing to previous valuable set of features. The measure is based on comparing of statistical distributions of individual features including mutual correlation. A mathematical theory of the approach is described. Results of the method applied to real-life data are shown.