View in EDS

Exponential distribution optimizer for improving multiple strategies in feature selection.

Saved in:
Bibliographic Details
Title: Exponential distribution optimizer for improving multiple strategies in feature selection.
Authors: Chen, Yufeng1 (AUTHOR) 22211860105@stu.wzu.edu.cn, Wang, Jian2 (AUTHOR) kenyoncy2016@wzu.edu.cn, Chen, Yi1 (AUTHOR), Heidari, Ali Asghar3 (AUTHOR) 23451350032@stu.wzu.edu.cn, Liu, Lei4 (AUTHOR), Wang, Mingjing5 (AUTHOR) liulei_cx@stu.scu.edu.cn, Chen, Huiling1 (AUTHOR) chenhuiling_jsj@wzu.edu.cn
Source: Cluster Computing. Dec2025, Vol. 28 Issue 15, p1-32. 32p.
Subject Terms: *FEATURE selection, *HEURISTIC algorithms, *MATHEMATICAL optimization, *CONSTRAINED optimization, *OPTIMIZATION algorithms
Abstract: As an emerging heuristic algorithm, the Exponential Distribution Optimizer (EDO) has gained widespread attention for its ability to explore and utilize search spaces. In high-dimensional optimization problems, algorithms often struggle with local optima, which hinders their ability to thoroughly explore the entire search space and limits their effectiveness in tackling complex challenges. This paper introduces an enhanced version of the exponential distribution optimizer (CDEDO) to overcome these limitations. The proposed improvements include the integration of horizontal and vertical crossover mechanisms (CC) and adaptive differential evolution mechanisms (ADE) to boost the algorithm's performance significantly. In order to verify the effectiveness of CDEDO, this paper conducted multiple experiments, particularly comparing it with traditional heuristic algorithms such as PSO, DE, and ABC, as well as improved algorithms such as SPS_L_SHADE_EIG, EBOwithCMAR, LSHADE_cnEpSi, SHADE, and BLPSO, on the IEEE CEC2017 benchmark test problem set. The experimental results show that CDEDO significantly improves convergence speed and exhibits a stronger ability to escape local optima. This indicates that CDEDO can better balance global and local search by introducing CC and ADE, thereby achieving better results in complex optimization problems. This study also applies CDEDO to high-dimensional feature selection problems. We selected 16 public real-world datasets. The experimental results show that CDEDO performs equally well on these datasets, effectively selecting feature subsets that are close to optimal, and outperforms other comparative algorithms on multiple datasets. By incorporating CC and ADE, CDEDO successfully overcomes the limitations of the original EDO, showcasing enhanced global search capabilities and faster convergence. [ABSTRACT FROM AUTHOR]
Database: Academic Search Index
Description
Abstract:As an emerging heuristic algorithm, the Exponential Distribution Optimizer (EDO) has gained widespread attention for its ability to explore and utilize search spaces. In high-dimensional optimization problems, algorithms often struggle with local optima, which hinders their ability to thoroughly explore the entire search space and limits their effectiveness in tackling complex challenges. This paper introduces an enhanced version of the exponential distribution optimizer (CDEDO) to overcome these limitations. The proposed improvements include the integration of horizontal and vertical crossover mechanisms (CC) and adaptive differential evolution mechanisms (ADE) to boost the algorithm's performance significantly. In order to verify the effectiveness of CDEDO, this paper conducted multiple experiments, particularly comparing it with traditional heuristic algorithms such as PSO, DE, and ABC, as well as improved algorithms such as SPS_L_SHADE_EIG, EBOwithCMAR, LSHADE_cnEpSi, SHADE, and BLPSO, on the IEEE CEC2017 benchmark test problem set. The experimental results show that CDEDO significantly improves convergence speed and exhibits a stronger ability to escape local optima. This indicates that CDEDO can better balance global and local search by introducing CC and ADE, thereby achieving better results in complex optimization problems. This study also applies CDEDO to high-dimensional feature selection problems. We selected 16 public real-world datasets. The experimental results show that CDEDO performs equally well on these datasets, effectively selecting feature subsets that are close to optimal, and outperforms other comparative algorithms on multiple datasets. By incorporating CC and ADE, CDEDO successfully overcomes the limitations of the original EDO, showcasing enhanced global search capabilities and faster convergence. [ABSTRACT FROM AUTHOR]
ISSN:13867857
DOI:10.1007/s10586-025-05686-2