Open AccessMethods for adapting the leaping frog algorithm to the binary search space when solving the feature selection problem
Author(s) -
Marina Bardamova,
A. G. Buymov,
V. F. Tarasenko
Publication year - 2020
Publication title -
doklady tomskogo gosudarstvennogo universiteta sistem upravleniâ i radioèlektroniki
Language(s) - English
Resource type - Journals
ISSN - 1818-0442
DOI - 10.21293/1818-0442-2020-23-4-57-62
Subject(s) - merge (version control) , metaheuristic , computer science , binary number , artificial intelligence , feature selection , pattern recognition (psychology) , classifier (uml) , algorithm , machine learning , mathematics , arithmetic , information retrieval
The feature selection is an important step in constructing any classifier. Binary versions of metaheuristic optimization algorithms are often used for selection. However, many metaheuristics are originally created to work in the continuous search space, so they need to be specially adapted to the binary space. In this paper, the authors propose fifteen ways to binarize the Shuffled frog leaping algorithm based on the following methods: modified algebraic operations, merge operation, and transformation functions. The efficiency of the binary algorithm was tested in the problem of feature selection for fuzzy classifiers on data sets from the KEEL repository. The results show that all the described methods of binarization allow reducing the features, while increasing the overall accuracy of classification.