<p>This paper proposes a novel multiobjective version based on Warlus optimizer (WO) called MOWO to optimize engineering, multi-objective optimization problems. The new MOWO utilize the update process in WO to evolve the position of candidate solutions. MOWO begins with Parent population initiated randomly, evolve the solutions with traditional WO which are siblings, optimal siblings are upgraded to Population. Archive matrix used to keep optimal solutions, distribution between Parent and siblings populations to strengthen information share and upgrade remain solutions relating to Pareto ones, furthermore, the proposed mutate-leaders strategy aims to improve the diversity of the obtained Pareto solutions and replace the centralized distribution of Pareto with random selection strategy so that diminish the risk of falling into local minima. The efficacy of the MOWO is assessed on a wide range of challenge optimization benchmarks, i.e., 20 test methods of multi-objective CEC’20 benchmark, optimize constrained ZDT benchmarks, and optimize the human activity recognition as multi-objective feature selection problem. Further, the MOWO results are compared to those obtained with seven other recent and well-known multiobjective methods. Various quantitative and qualitative metrics are employed to conduct a comprehensive examination of the results. On the activity recognition side, two multiobjective feature selection variants are introduced <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10586_2025_5340_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="104" /> </InlineMediaObject> <EquationSource Format="TEX">$bMOWO_{KNN}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>b</mi> <mi>M</mi> <mi>O</mi> <mi>W</mi> <msub> <mi>O</mi> <mrow> <mi mathvariant="italic">KNN</mi> </mrow> </msub> </mrow> </math></EquationSource> </InlineEquation> and <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10586_2025_5340_Article_IEq2.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="90" /> </InlineMediaObject> <EquationSource Format="TEX">$bMOWO_{RF}$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>b</mi> <mi>M</mi> <mi>O</mi> <mi>W</mi> <msub> <mi>O</mi> <mrow> <mi mathvariant="italic">RF</mi> </mrow> </msub> </mrow> </math></EquationSource> </InlineEquation>, that based on MOWO and employing the binary transfer function, classification results are compared to several literature studies, two public action recognition datasets i.e. UniMib-SHAR, Opportunity are employed to validate the introduced variants with accuracy 86.60% and 87.3% respectively, superior to literature studies. The experimental results demonstrate that the MOWO algorithm is reliable and trustworthy in determining the optimal region, besides a proper convergence rate towards the near-optimal regions when compared to competitors. Indeed, the MOWO demonstrates superior performance comparing to other competitors for majority of the benchmarks.</p>

An efficient Warlus optimizer for solving multi-objective optimization problems: case study with action recognition

期刊论文
出版模式：

开放获取
发表日期：

2025年9月3日

查看全文

Essam H. Houssein,
Mohamed A. Mahdy,
Mohammed Kayed,
Waleed M. Mohamed

摘要

This paper proposes a novel multiobjective version based on Warlus optimizer (WO) called MOWO to optimize engineering, multi-objective optimization problems. The new MOWO utilize the update process in WO to evolve the position of candidate solutions. MOWO begins with Parent population initiated randomly, evolve the solutions with traditional WO which are siblings, optimal siblings are upgraded to Population. Archive matrix used to keep optimal solutions, distribution between Parent and siblings populations to strengthen information share and upgrade remain solutions relating to Pareto ones, furthermore, the proposed mutate-leaders strategy aims to improve the diversity of the obtained Pareto solutions and replace the centralized distribution of Pareto with random selection strategy so that diminish the risk of falling into local minima. The efficacy of the MOWO is assessed on a wide range of challenge optimization benchmarks, i.e., 20 test methods of multi-objective CEC’20 benchmark, optimize constrained ZDT benchmarks, and optimize the human activity recognition as multi-objective feature selection problem. Further, the MOWO results are compared to those obtained with seven other recent and well-known multiobjective methods. Various quantitative and qualitative metrics are employed to conduct a comprehensive examination of the results. On the activity recognition side, two multiobjective feature selection variants are introduced $bMOWO_{KNN}$ $b M O W O_{KNN}$ and $bMOWO_{RF}$ $b M O W O_{RF}$ , that based on MOWO and employing the binary transfer function, classification results are compared to several literature studies, two public action recognition datasets i.e. UniMib-SHAR, Opportunity are employed to validate the introduced variants with accuracy 86.60% and 87.3% respectively, superior to literature studies. The experimental results demonstrate that the MOWO algorithm is reliable and trustworthy in determining the optimal region, besides a proper convergence rate towards the near-optimal regions when compared to competitors. Indeed, the MOWO demonstrates superior performance comparing to other competitors for majority of the benchmarks.