GENETIC ALGORITHM BASED FEATURE SELECTION FOR EXTREME LEARNING MACHINES

Full Article - PDF

Published: 2016-07-11

Page: 34-39


SARTHAK YADAV

Krishna Institute of Engineering and Technology (KIET), Ghaziabad, India.

ANKUR SINGH BIST *

Krishna Institute of Engineering and Technology (KIET), Ghaziabad, India.

*Author to whom correspondence should be addressed.


Abstract

Extreme Learning Machines are a novel method for classification, which builds upon the working methodology of Single Layer Feed Forward Networks and tries to improve the training speed, while still striving for a comparable or better Classification performance.

But in practical classification and pattern recognition, many times the data we need to work upon possesses undesirable qualities like high dimensionality, duplicated and non-related features and missing/null values. Thus there is a requirement of selection of a subset of attributes or features to represent the patterns to be classified.

To deal with these undesirable qualities, we need to preprocess data and transform it into a more usable form. This process is called Feature Selection, i.e. processing and applying various transforms that converts datasets into a much more usable and desirable form.

In this work, we apply Genetic Algorithm along with various well known feature selection methodologies such as Selecting K Best Features and Selecting a Percentile of features, using ANOVA F-value and Chi2 Scoring to search out and identify the potential informative feature combinations for classification and then use Extreme Learning Machines to compare and contrast the results obtained from various Feature Selection algorithms used and thus demonstrate the viability of Genetic Algorithm for feature selection.

Keywords: ELM, classifiers, classification, feature selection, genetic algorithms, ANOVA F-value, Chi2, SelectKBest, SelectPercentile


How to Cite

YADAV, S., & BIST, A. S. (2016). GENETIC ALGORITHM BASED FEATURE SELECTION FOR EXTREME LEARNING MACHINES. Asian Journal of Mathematics and Computer Research, 13(1), 34–39. Retrieved from https://ikprress.org/index.php/AJOMCOR/article/view/686

Downloads

Download data is not yet available.