Scholarlink

<p>Data clustering is a prominent unsupervised learning method that continues to attract significant research interest due to its diverse applications across various domains. Partitional clustering methods like K-Means optimize a criterion function to identify homogeneous groups in a dataset. However, random initialization can lead to non-optimal cluster centroids. While combining K-Means with metaheuristics improves performance in data analysis tasks, the basic learning strategies, decline in diversity and inadequate exploration and exploitation in population-based metaheuristic algorithms still cause premature convergence of cluster centroids to local optima. To address this challenge, this work proposes the hybrid opposition-based improved particle swarm optimization algorithm (HOB-iPSO) for partitional data clustering. HOB-iPSO employs a hybrid approach for initializing cluster centroids by integrating random estimation, quasi-opposition-based learning, and K-Means to improve exploration capability, maintain diversity, and identify various promising regions in the high-dimensional feature space. Moreover, HOB-iPSO achieves a balanced exploration–exploitation using logistic map-based chaotic inertia weight, promoting stable convergence of cluster centroids for improved data clustering solutions. The effectiveness of HOB-iPSO is evaluated on thirteen real-world datasets from the UCI Machine Learning Repository, and the results obtained with HOB-iPSO are compared with K-Means and existing metaheuristic-based data clustering algorithms using both internal and external validation metrics. The proposed HOB-iPSO improves the performance of data clustering by reducing the sum of intra-cluster distances and the Davies–Bouldin index and by increasing the silhouette coefficient, F-measure, and accuracy, compared to the other data clustering algorithms. Statistical significance is evaluated using the Friedman test, followed by Wilcoxon signed-rank post hoc tests with Holm correction. The numerical results and graphical visualizations, including ablation studies, runtime comparison for practical applicability, convergence curve analysis, and 3D scatter plots, confirm that the HOB-iPSO is reliable and effective in producing well-formed clusters.</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

Well-formed partitional data clusters using hybrid opposition-based improved particle swarm optimization

  • 期刊论文

  • 发表日期：

    2026年5月19日

查看全文
  • Jaswinder Singh,
  • Damanpreet Singh

摘要

Data clustering is a prominent unsupervised learning method that continues to attract significant research interest due to its diverse applications across various domains. Partitional clustering methods like K-Means optimize a criterion function to identify homogeneous groups in a dataset. However, random initialization can lead to non-optimal cluster centroids. While combining K-Means with metaheuristics improves performance in data analysis tasks, the basic learning strategies, decline in diversity and inadequate exploration and exploitation in population-based metaheuristic algorithms still cause premature convergence of cluster centroids to local optima. To address this challenge, this work proposes the hybrid opposition-based improved particle swarm optimization algorithm (HOB-iPSO) for partitional data clustering. HOB-iPSO employs a hybrid approach for initializing cluster centroids by integrating random estimation, quasi-opposition-based learning, and K-Means to improve exploration capability, maintain diversity, and identify various promising regions in the high-dimensional feature space. Moreover, HOB-iPSO achieves a balanced exploration–exploitation using logistic map-based chaotic inertia weight, promoting stable convergence of cluster centroids for improved data clustering solutions. The effectiveness of HOB-iPSO is evaluated on thirteen real-world datasets from the UCI Machine Learning Repository, and the results obtained with HOB-iPSO are compared with K-Means and existing metaheuristic-based data clustering algorithms using both internal and external validation metrics. The proposed HOB-iPSO improves the performance of data clustering by reducing the sum of intra-cluster distances and the Davies–Bouldin index and by increasing the silhouette coefficient, F-measure, and accuracy, compared to the other data clustering algorithms. Statistical significance is evaluated using the Friedman test, followed by Wilcoxon signed-rank post hoc tests with Holm correction. The numerical results and graphical visualizations, including ablation studies, runtime comparison for practical applicability, convergence curve analysis, and 3D scatter plots, confirm that the HOB-iPSO is reliable and effective in producing well-formed clusters.