Elite-Guided Nutcracker Optimization Algorithm with Memory-Differential Spiral Search for Feature Selection

Baowei Pang

doi:10.3390/biomimetics11100684

Elite-Guided Nutcracker Optimization Algorithm with Memory-Differential Spiral Search for Feature Selection

2026-09-23

0

OA

AI

DOI：10.3390/biomimetics11100684

Original

Original request for help

Save

Abstract

En 中文

Feature selection can effectively remove redundant and irrelevant features in classification tasks; however, feature subset search on datasets with different dimensionalities still faces challenges such as strong initialization randomness, complex search spaces, a single local exploitation direction, and difficulty in effectively updating inferior individuals. This paper proposes an Elite-Guided Nutcracker Optimization Algorithm with Memory-Differential Spiral Search for Feature Selection (EGRNOA). The proposed method first employs Minimum Redundancy Maximum Relevance (mRMR) to preselect candidate features, thereby reducing the interference of redundant features in the subsequent search process. Then, Logistic–Tent chaotic initialization is used to enhance the uniformity of the initial population distribution. In the foraging phase of NOA, a memory-differential spiral search strategy is designed in which global best memory, individual historical best memory, and random differential directions are jointly incorporated into the spiral updating process so that the search no longer contracts only around a single optimal individual, thereby enhancing local exploitation capability while maintaining directional perturbation. A diversity-adaptive elite-guided reconstruction mechanism for inferior individuals is constructed; this mechanism uses elite population information to directionally repair inferior individuals, which improves population update efficiency while alleviating premature convergence. Experimental results on the CEC2020 benchmark functions show that EGRNOA achieves the best or tied-best average results on all ten functions and obtains the best Friedman average ranking of 1.050. On ten classification datasets with different dimensionalities, EGRNOA achieves the highest average classification accuracy on nine; across datasets, its average accuracy on the movement dataset is improved by 6.41 percentage points when compared with the second-best algorithm. Convergence curves, box plots, critical difference diagrams, and ablation experiments further verify the effectiveness of the memory-differential spiral search and diversity-adaptive elite-guided reconstruction mechanism in improving search quality and enhancing the robustness of feature selection.

Keywords:

feature selection

multidimensional datasets

nutcracker optimization algorithm

mRMR

logistic–tent chaotic mapping

memory-differential spiral search

elite-guided search

Elite-Guided Nutcracker Optimization Algorithm with Memory-Differential Spiral Search for Feature Selection

Abstract

Journal

Organization

Cited Papers