Pareto-Guided Leader Selection in the Mountain Gazelle Optimizer: A Controlled Multi-Objective Study
- 1Department of Computer Science, Faculty of Informatics, Mahasarakham University, Maha Sarakham 44150, Thailand
Abstract
Where should multi-objective reasoning enter a swarm optimizer: before the search through scalarization, around the search through survival or archiving, or inside the search through Pareto-informed leader selection? We study this question on the Mountain Gazelle Optimizer (MGO) and propose a feedback-path taxonomy that separates the three roles. A census of twenty multi-objective MGO extensions finds one prior leader-mediated design and no method that assigns per-agent leaders from a Pareto archive. We develop four leader-mediated variants on one search skeleton, with a random-leader control and a leader-isolation ablation. On twenty-two benchmark problems (thirty runs each), all four variants rank ahead of five paper-based reconstructions of prior MGO designs on three indicators, and the decomposition and indicator leader rules outrank random leadership on hypervolume and IGD+ under Holm correction. On fourteen held-out engineering problems, a refined indicator-guided design separates from NSGA-II, NSGA-III, MOEA/D, AGE-MOEA-II and SMS-EMOA on IGD+ under returned-set scoring, and from all but SMS-EMOA on hypervolume; under evaluation-trace scoring, only from NSGA-III and MOEA/D on IGD+. In sample, of the five it separates only from MOEA/D (ε-indicator, returned-set scoring). The four variants outperform the prior MGO designs as complete implementations; against established algorithms, the advantage depends on scoring.
Related
- Nash-Bargaining HalpernSGD via Limited-Memory MultiLRSGA: A Two-Phase Optimizer for Multi-Objective Learning
- Objective Bayes Factor Based Model Selection for Zero-Inflated Microbiome Count Data Under Jeffreys Prior
- Objective Selection in Many-Objective Optimization via Conditioning-Based Analysis of Objective Redundancy
- Model selection in linear mixed-effects analysis of craving outcomes in a randomized controlled trial of medications for opioid use disorders
- FDR-Controlled Variable Selection for Generalized Linear Models and Cox Regression with Virtual Dummies
BibTeX
@article{thanarat2026paretoguided,
title = {{Pareto-Guided Leader Selection in the Mountain Gazelle Optimizer: A Controlled Multi-Objective Study}},
author = {Preut Thanarat and Suchart Khummanee and Chatklaw Jareanpon},
journal = {Mathematics},
year = {2026},
month = nov,
doi = {10.3390/math14214043},
url = {https://www.mdpi.com/2227-7390/14/21/4043},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Pareto-Guided Leader Selection in the Mountain Gazelle Optimizer: A Controlled Multi-Objective Study
AU - Preut Thanarat
AU - Suchart Khummanee
AU - Chatklaw Jareanpon
PY - 2026
DA - 2026/11/04
JO - Mathematics
DO - 10.3390/math14214043
UR - https://www.mdpi.com/2227-7390/14/21/4043
AB - Where should multi-objective reasoning enter a swarm optimizer: before the search through scalarization, around the search through survival or archiving, or inside the search through Pareto-informed leader selection? We study this question on the Mountain Gazelle Optimizer (MGO) and propose a feedback-path taxonomy that separates the three roles. A census of twenty multi-objective MGO extensions finds one prior leader-mediated design and no method that assigns per-agent leaders from a Pareto archive. We develop four leader-mediated variants on one search skeleton, with a random-leader control and a leader-isolation ablation. On twenty-two benchmark problems (thirty runs each), all four variants rank ahead of five paper-based reconstructions of prior MGO designs on three indicators, and the decomposition and indicator leader rules outrank random leadership on hypervolume and IGD+ under Holm correction. On fourteen held-out engineering problems, a refined indicator-guided design separates from NSGA-II, NSGA-III, MOEA/D, AGE-MOEA-II and SMS-EMOA on IGD+ under returned-set scoring, and from all but SMS-EMOA on hypervolume; under evaluation-trace scoring, only from NSGA-III and MOEA/D on IGD+. In sample, of the five it separates only from MOEA/D (ε-indicator, returned-set scoring). The four variants outperform the prior MGO designs as complete implementations; against established algorithms, the advantage depends on scoring.
SN - 2227-7390
ER -