Simplifying pattern recognition problems via a scatter search algorithm
Antonio J. Tallón‐Ballesteros, Alberto Ibiza-Granados · International Journal for Computational Methods in Engineering Science and Mechanics · 2016
This article presents an approach to simplify pattern recognition problems via a scatter search algorithm that is applied as feature subset selector (FSS). Experimentation on five high-dimensionality problems, with a feature space in the range 2308–16063 and feature-to-pattern ratios greater than 27, revealed that the most appropriate feature selector is based on correlation. Moreover, the most accurate way is to combine the new proposal with a correlation-based attribute evaluator with a Naive Bayes Tree classifier; their performance has been compared with a reference FSS and sheds light on very interesting results in terms of accuracy and problem reduction.