Novel binary walrus optimization algorithms BWaOA and BWaOA-C with crossover operator for feature selection in high-dimensional data

Abstract Redundant and irrelevant features in high-dimensional datasets hinder the development of efficient machine learning models. Most existing Feature Selection (FS) algorithms are developed based on either embedded or filter techniques, which makes it challenging to identify the highly discrimi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Discover Computing Jg. 28; H. 1; S. 1 - 45
Hauptverfasser: Farid Ayeche, Adel Alti
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Springer 27.10.2025
Schlagworte:
Binary walrus optimization algorithm
Classification
Feature selection
K-Nearest neighbor
Machine learning
ISSN:2948-2992
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Abstract Redundant and irrelevant features in high-dimensional datasets hinder the development of efficient machine learning models. Most existing Feature Selection (FS) algorithms are developed based on either embedded or filter techniques, which makes it challenging to identify the highly discriminant features due to limited search capability and high computational cost. To overcome these challenges, we propose a novel wrapper-based FS framework built on the Walrus Optimization Algorithm (WaOA) to balance accuracy and efficiency. The key novelties of our framework include two advanced binarization strategies: Binary WaOA (BWaOA), which uses S- and V-shaped transfer functions for effective search space discretization, and Binary WaOA-Crossover (BWaOA-C), which incorporates crossover operators to improve exploration, diversity, and refinement. Unlike conventional approaches, our methods systematically combine adaptive transfer functions and dynamic thresholding to select compact yet highly discriminative feature subsets, evaluated using a K-Nearest Neighbors (KNN) classifier. Extensive experiments on 30 benchmark datasets demonstrate the superiority of the proposed framework against 12 state-of-the-art FS algorithms, including GA, PSO, HHO, GWO, ChOA, BDE, WOA, AMGWO, BTLBO-KNN, HLBDA, BABC, and RGA-T. BWaOA achieves 86.33% feature reduction and 92.56% classification accuracy, while BWaOA-C further improves accuracy by up to 7%. These findings demonstrate the robustness and practical effectiveness of the proposed framework for high-dimensional data analysis.
ISSN:2948-2992
DOI:10.1007/s10791-025-09767-z