A novel binary puma optimization algorithm for feature selection problem

Sayed Zabihullah Musawi

doi:10.1007/s10586-026-06610-y

Return

A novel binary puma optimization algorithm for feature selection problem

2026-09-17

PRE

Sayed Zabihullah Musawi

Mohammad Farshi *

Sepehr Ebrahimi Mood

DOI：10.1007/s10586-026-06610-y

Original

Original request for help

Save

Abstract

En 中文

The Puma Optimization Algorithm (POA) is a recently proposed metaheuristic inspired by the cooperative hunting strategies and social behavior of pumas. This study develops a Binary POA (BPOA) specifically designed for feature selection by integrating ten Transfer Functions (TFs). These transformations enable POA to operate effectively within binary search spaces, thereby extending its applicability to high-dimensional classification problems. The evaluation procedure comprised three stages: (i) identifying the most suitable TF for BPOA, (ii) benchmarking against conventional feature selection methods, and (iii) by comparing performance with leading metaheuristic algorithms, including the Binary Slime Mould Algorithm (BSMA), Binary Harris Hawks Optimization (BHHO), Binary Flood Algorithm (BFLA), Binary Chinese Pangolin Optimizer (BCPO), and Binary Rüppell’s Fox Optimizer (BRFO). Performance was assessed through classification accuracy, fitness value, number of selected features, and computational time. K-Nearest Neighbour (KNN) was used as the primary classifier, complemented by Support Vector Machine (SVM), Gaussian Naïve Bayes (GNB), and Logistic Regression (LGR) for further analysis. Results indicate that the Threshold Cut-off variant of BPOA (BPOATC) achieves the highest classification accuracy (91.53%) among all TF-based versions and consistently demonstrates superior performance compared to both traditional feature selection techniques and state-of-the-art metaheuristics. Friedman test outcomes further confirm its statistical superiority. Additional validation on three Intrusion Detection System (IDS) datasets (NSL-KDD, UNSW-NB15, UAV-IDS-2020) demonstrates strong robustness, with BPOATC attaining leading performance on UNSW-NB15, and UAV-IDS-2020 and near-optimal results on NSL-KDD.

Keywords:

Machine learning

Feature selection

Optimization algorithms

Puma optimization algorithm

Intrusion detection system