Scholarlink

<p>This study focuses on developing a novel approach for clustering high-dimensional sparse binary data with a focus on product categorization in the information sector. The proposed method combines Binary Dimension Reduction (BDR) with a Genetic Algorithm (GA) and Fuzzy K-Modes (FKM) clustering approach, and refers to it as BDR-GAFKM. BDR reduces the dimensionality while maintaining binary semantics to handle data sparsity and complexity. To overcome FKM’s dependence on random initialization and its tendency to converge to local optima, GA is incorporated to optimize cluster centroids globally. A new Collision crossover and mutation method is investigated to improve GA performance on binary datasets. Besides, the proposed BDR-GAKFM employs both within-cluster variance and maximum entropy in its objective function to improve clustering compactness and separation. The computational findings on twelve benchmark datasets reveal that the proposed BDR-GAFKM significantly outperforms existing algorithms in terms of accuracy, precision, recall, and <i>F</i>1-score. Moreover, the combination of the entropy term in the objective function also enhances clustering performance, resulting in an improvement of 10.59% in accuracy, 12.71% in recall, 11.41% in precision, and 12.89% in F1-score. Finally, a practical case study validates the efficiency of the proposed BDR-GAFKM in reducing production line setup time through improved product clustering.</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

Product clustering using dimensional reduction and GA-based clustering algorithm for the information industry

  • 期刊论文

  • 发表日期：

    2026年3月14日

查看全文
  • R. J. Kuo,
  • Chia-Jung Fan,
  • Thi Phuong Quyen Nguyen,
  • C.-W. Shih

摘要

This study focuses on developing a novel approach for clustering high-dimensional sparse binary data with a focus on product categorization in the information sector. The proposed method combines Binary Dimension Reduction (BDR) with a Genetic Algorithm (GA) and Fuzzy K-Modes (FKM) clustering approach, and refers to it as BDR-GAFKM. BDR reduces the dimensionality while maintaining binary semantics to handle data sparsity and complexity. To overcome FKM’s dependence on random initialization and its tendency to converge to local optima, GA is incorporated to optimize cluster centroids globally. A new Collision crossover and mutation method is investigated to improve GA performance on binary datasets. Besides, the proposed BDR-GAKFM employs both within-cluster variance and maximum entropy in its objective function to improve clustering compactness and separation. The computational findings on twelve benchmark datasets reveal that the proposed BDR-GAFKM significantly outperforms existing algorithms in terms of accuracy, precision, recall, and F1-score. Moreover, the combination of the entropy term in the objective function also enhances clustering performance, resulting in an improvement of 10.59% in accuracy, 12.71% in recall, 11.41% in precision, and 12.89% in F1-score. Finally, a practical case study validates the efficiency of the proposed BDR-GAFKM in reducing production line setup time through improved product clustering.