Evaluation of Filter and Wrapper Methods for Feature Selection in Supervised Machine Learning

NONSO NNAMOKO, F Arshad, D England, J Vora, J Norman

Research output: Chapter in Book/Report/Conference proceedingConference proceeding (ISBN)peer-review


Feature selection methods for classification is a
subject often addressed in Machine Learning tasks; particularly
where there are many features with comparatively few samples.
However, their performance depends on a number of factors
such as data, classifier etc. This paper, examines adapting an
existing wrapper and filter method to perform experiments with
a diabetes diagnosis dataset (768 x 9) from UCI database. Two
classifiers (NaïveBayes and Bagging) were adopted to compare
performance of the selection approaches, using Accuracy and
Receiver Operating Characteristics area (aROC) as matric.
Result confirms the utility of feature selection for classification
and the superiority of wrapper methods. However, some
problems do arise from using wrapper methods and, evidence is
proposed that filters are a reasonable alternative with limited
computational cost for dealing with large datasets.
Original languageEnglish
Title of host publicationPGNET Proceedings of the 15th Annual Postgraduate Symposium on the Convergence of Telecommunications, Networking and Broadcasting 2014
EditorsMadjid Merabti, Omar Abuelma'atti, Carol Oliver
PublisherLiverpool John Moores University, School of Computing & Mathematical Sciences
ISBN (Electronic)9781902560274
ISBN (Print)9781902560281
Publication statusPublished - 23 Jun 2014


  • machine learning
  • classification
  • wrapper
  • filter


Dive into the research topics of 'Evaluation of Filter and Wrapper Methods for Feature Selection in Supervised Machine Learning'. Together they form a unique fingerprint.

Cite this