Feature Selection From Partially Uncertain Data Within the Belief Function Framework
Résumé
With the rapid growth of high dimensional data, feature selection has become a substantial task for several machine learning problems. In fact, it is regarded as an important process for classification performance owing to its ability to remove redundant and inconsistent features. The rough set theory is regarded as a well known tool allowing relevant feature selection. As the task of attribute selection using rough sets is an NP-hard problem, several heuristic algorithms have been introduced. The Johnson's algorithm, handling data characterized by certain and precise attribute values, is one of the most known ones. In this paper, we propose to extend this latter algorithm to an uncertain context, precisely where data contain uncertain condition attribute values represented within the belief function framework. We test the performance of our belief Johnson's algorithm through several experiments on synthetic databases.
Origine | Fichiers produits par l'(les) auteur(s) |
---|