Scholarlink

In practical applications, data stream classification faces significant challenges, such as the high cost of data labeling and class imbalance. The former is time-consuming and labor-intensive, while the latter may lead online learning models to favor majority class samples, resulting in incorrect decisions for minority class samples. To address these issues, this paper proposes Density Harmonized Gradient Descent Online Active Learning (DHGDOAL). This method first computes the local density factor of samples based on a sliding window and prioritizes instances located in high-density regions for labeling, thereby annotating valuable samples with a limited query budget. Second, to avoid the model from exhibiting a bias towards the majority class during update, we introduce a gradient harmonization mechanism. Finally, experimental results show that DHGDOAL has high performance in terms of accuracy, AUC, F1-score, and G-MEANS when dealing with imbalanced data streams under a limited budget.

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

Density Harmonized Gradient Descent Online Active Learning for Imbalanced Data Streams

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jianhao Huang,
  • Zhihong Zeng,
  • Zhangkai Wu,
  • Hengyu Liu,
  • Yanfang Liu

摘要

In practical applications, data stream classification faces significant challenges, such as the high cost of data labeling and class imbalance. The former is time-consuming and labor-intensive, while the latter may lead online learning models to favor majority class samples, resulting in incorrect decisions for minority class samples. To address these issues, this paper proposes Density Harmonized Gradient Descent Online Active Learning (DHGDOAL). This method first computes the local density factor of samples based on a sliding window and prioritizes instances located in high-density regions for labeling, thereby annotating valuable samples with a limited query budget. Second, to avoid the model from exhibiting a bias towards the majority class during update, we introduce a gradient harmonization mechanism. Finally, experimental results show that DHGDOAL has high performance in terms of accuracy, AUC, F1-score, and G-MEANS when dealing with imbalanced data streams under a limited budget.