Scholarlink

<p>In the minimum sum-of-squares clustering problem, several local optima often cause solutions to stagnate because it is a nonlinear and nonconvex optimization problem. In this paper, we propose a hybrid algorithm, GA-TS, which combines the global search capability of the genetic algorithm (GA) with the local optimization efficiency of tabu search (TS). As a result of integrating complementary algorithms and using an opposition-based method to initialize the population, GA-TS consistently produces higher-quality results. Results from 23 datasets validate its effectiveness, with GA-TS achieving the best results in 21 datasets. On the Glass Identification dataset, GA-TS achieved a best objective value of 18.2412, significantly outperforming GA (54.4705) and TS (27.1996). GA-TS achieved 416.5171 on the Dermatology dataset, exceeding GA (999.0877) and TS (471.8221). Similarly, GA-TS delivered a best value of 2810.8350 on the BreastB dataset, outperforming GA (3236.7023) and the Differential Evolution (DE) (2815.4374). According to the BreastA dataset, GA-TS recorded 3081.4873, which is better than TS (3110.7432) and DE (3142.2720). It is noteworthy that GA-TS performed better than GA (38.6569) and TS (7.3609) for the Iris dataset in all ten runs. In comparison with methods like Differential Evolution, Simulated Annealing (SA), and K-means, GA-TS consistently minimized both the best and worst objective values. Under the minimum sum-of-squares criterion, GA-TS offers a robust and scalable solution for diverse and challenging optimization problems, underscoring its efficiency and effectiveness.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A dual-phase strategy for clustering: integrating genetic algorithms with tabu search

  • 期刊论文

  • 发表日期：

    2025年9月23日

查看全文
  • Najme Nazari,
  • Mohammad Ali Yaghoobi,
  • Najme Mansouri

摘要

In the minimum sum-of-squares clustering problem, several local optima often cause solutions to stagnate because it is a nonlinear and nonconvex optimization problem. In this paper, we propose a hybrid algorithm, GA-TS, which combines the global search capability of the genetic algorithm (GA) with the local optimization efficiency of tabu search (TS). As a result of integrating complementary algorithms and using an opposition-based method to initialize the population, GA-TS consistently produces higher-quality results. Results from 23 datasets validate its effectiveness, with GA-TS achieving the best results in 21 datasets. On the Glass Identification dataset, GA-TS achieved a best objective value of 18.2412, significantly outperforming GA (54.4705) and TS (27.1996). GA-TS achieved 416.5171 on the Dermatology dataset, exceeding GA (999.0877) and TS (471.8221). Similarly, GA-TS delivered a best value of 2810.8350 on the BreastB dataset, outperforming GA (3236.7023) and the Differential Evolution (DE) (2815.4374). According to the BreastA dataset, GA-TS recorded 3081.4873, which is better than TS (3110.7432) and DE (3142.2720). It is noteworthy that GA-TS performed better than GA (38.6569) and TS (7.3609) for the Iris dataset in all ten runs. In comparison with methods like Differential Evolution, Simulated Annealing (SA), and K-means, GA-TS consistently minimized both the best and worst objective values. Under the minimum sum-of-squares criterion, GA-TS offers a robust and scalable solution for diverse and challenging optimization problems, underscoring its efficiency and effectiveness.