Scholarlink

<p>Imbalanced data classification remains a significant challenge in machine learning, as conventional learning algorithms are typically biased toward majority classes, often leading to poor recognition of minority instances. This limitation is particularly critical in binary classification tasks, where correct identification of the minority class plays a crucial role in decision making. Consequently, developing effective strategies to mitigate class imbalance has become an important research direction in data mining and knowledge discovery. To address this issue, this study proposes a data-level solution: a metaheuristic-based rebalancing framework. The approach leverages metaheuristic optimization techniques to produce a more balanced data distribution by increasing the representation of minority-class instances while preserving the intrinsic characteristics and structure of the original dataset. Experiments using 5Cv-fold cross-validation on benchmark datasets with varying imbalance ratios (IR ≤ 3, 3 &lt; IR ≤ 9, IR&gt;9) demonstrate that the proposed rebalancing method substantially improves classifier performance across several common evaluation metrics. The results show consistent gains over the baseline, with the F1-score improving (≈&#xa0;35.3%), recall increasing (≈&#xa0;47.6%), and G-mean improving (≈&#xa0;21.5%). These results indicate that the optimized datasets enable classifiers to achieve better balance in overall metrics and significantly enhance detection of minority-class instances. Furthermore, statistical significance analysis confirms that the observed improvements are consistent and not due to random variation. Overall, the findings highlight the effectiveness of metaheuristic-driven data-level rebalancing strategies as a powerful tool for addressing class imbalance, improving model generalization, and strengthening minority-class recognition in challenging machine-learning scenarios.</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

Metaheuristic-based data-level rebalancing for imbalanced binary classification: an empirical study

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月7日

查看全文
  • Laouni Djafri

摘要

Imbalanced data classification remains a significant challenge in machine learning, as conventional learning algorithms are typically biased toward majority classes, often leading to poor recognition of minority instances. This limitation is particularly critical in binary classification tasks, where correct identification of the minority class plays a crucial role in decision making. Consequently, developing effective strategies to mitigate class imbalance has become an important research direction in data mining and knowledge discovery. To address this issue, this study proposes a data-level solution: a metaheuristic-based rebalancing framework. The approach leverages metaheuristic optimization techniques to produce a more balanced data distribution by increasing the representation of minority-class instances while preserving the intrinsic characteristics and structure of the original dataset. Experiments using 5Cv-fold cross-validation on benchmark datasets with varying imbalance ratios (IR ≤ 3, 3 < IR ≤ 9, IR>9) demonstrate that the proposed rebalancing method substantially improves classifier performance across several common evaluation metrics. The results show consistent gains over the baseline, with the F1-score improving (≈ 35.3%), recall increasing (≈ 47.6%), and G-mean improving (≈ 21.5%). These results indicate that the optimized datasets enable classifiers to achieve better balance in overall metrics and significantly enhance detection of minority-class instances. Furthermore, statistical significance analysis confirms that the observed improvements are consistent and not due to random variation. Overall, the findings highlight the effectiveness of metaheuristic-driven data-level rebalancing strategies as a powerful tool for addressing class imbalance, improving model generalization, and strengthening minority-class recognition in challenging machine-learning scenarios.