A Binary Water Flow Optimizer Applied to Feature Selection

Abstract

In this work, a binary version of the Water Flow Optimizer (WFO) algorithm, called Binary Water Flow Optimizer (BWFO), is introduced addressing the feature selection problem. WFO is an evolutionary algorithm inspired by the way water flows in nature. In this new approach, the BWFO uses the laminar flow and turbulent flow operators in a binary version, using the Optimum-Path Forest (OPF) classifier as a fitness function. The proposed approach is evaluated through a comparative analysis made with classical methods of dimensionality reduction, more specifically with the Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) and with the metaheuristics Binary Water Wave Optimization (BWWO), Binary Bat Algorithm (BBA) and Binary Cuckoo Search (BCS). The computational results demonstrate that the approach is a valid and effective alternative to the feature selection problem.