Praznik: High performance information-based feature selection

Abstract

Information filters are an important class of feature selection methods. They bring together strong theoretical background, execution speed and high selection quality. The paper provides a brief review of said methods and introduces praznik, an R package collecting their efficient implementations. Its functionality is discussed with illustrative example analyses on established benchmark datasets, which also demonstrate the package’s competitive accuracy and very high computational efficiency in comparison to other tools. Moreover, praznik exposes its low-level functionality, which aims to support research and development of novel information theory based methods.

Topics & Concepts