Scholarlink

<p>Learning non-stationary data streams faces significant challenges. This can be attributed to the high cost of labeling the unbalanced with potential concept drifts. The concept drift methodology is merely based on the errors or inconveniences that occur owing to inculcating new data or change in existing data streams. Block ensemble learning can tackle both improving global performance and buildings an intelligent model to perform classification on stream input. To deal with the class imbalance an enhanced particle swarm optimization based oversampling method is employed. The block based dynamic ensemble selection method evolves a base classifier on the current balanced data chunk and merges it with the previous classifiers to form a classifier pool. Per the test input, optimal grouping is performed in terms of the best neighborhood. An experimental analysis of five synthetic and four real-world datasets proves that the proposed technique outperforms the existing techniques in terms of classification predictive measures, such as G-mean and Area under Curve. The effectiveness of the approach is also tested using the Friedman test and posthoc analysis to prove its stability.</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

Block based ensemble learning on imbalance data stream with concept drift using enhanced particle swarm optimization

  • 期刊论文

  • 发表日期：

    2026年3月21日

查看全文
  • M. Blessa Binolin Pepsi,
  • N. Senthil Kumar

摘要

Learning non-stationary data streams faces significant challenges. This can be attributed to the high cost of labeling the unbalanced with potential concept drifts. The concept drift methodology is merely based on the errors or inconveniences that occur owing to inculcating new data or change in existing data streams. Block ensemble learning can tackle both improving global performance and buildings an intelligent model to perform classification on stream input. To deal with the class imbalance an enhanced particle swarm optimization based oversampling method is employed. The block based dynamic ensemble selection method evolves a base classifier on the current balanced data chunk and merges it with the previous classifiers to form a classifier pool. Per the test input, optimal grouping is performed in terms of the best neighborhood. An experimental analysis of five synthetic and four real-world datasets proves that the proposed technique outperforms the existing techniques in terms of classification predictive measures, such as G-mean and Area under Curve. The effectiveness of the approach is also tested using the Friedman test and posthoc analysis to prove its stability.