Feature Selection Based on Wrapper and Information Gain

Authors:
Publisher website
Google Scholar
Abstract
The accuracy of the classification process always suffers from the high dimensionality problem due to the independent, irrelevant, redundant and not useful attributes of the dataset. In this research, feature selection techniques (wrapper selection method, and information gain method) are obtained to handle the mentioned problem by removing those features and reducing the dataset dimensions. The techniques include wrapper selection method and information gain method. This research predicates on the diabetes dataset in WEKA application, which contains checking seven models enforce wrapper selection method as an attribute evaluator, forwarding direction, backward, and bi-directional best-first search method and Naïve Bayes technique as a classifier method, checking eight models applying information gain method as attribute evaluator, as well as the ranker as a search method. Additional to demonstrate the decision tree and classification figures for the best-obtained models in each one technique. The results proved the ability of wrapper and information gain to choose a minimum number of features in order to classify the data with an accuracy of more than 76% in this work.

This publication has 14 references indexed in Scilit: