Scholarlink

This paper introduces the Memory-based Streaming Classification for Anomaly Detection (MSCAD), which combines supervised memory updates with adaptive thresholding to enhance performance in streaming data environments. Evaluations were conducted on three publicly available datasets: KDD Cup 1999 (KDD99), NSL-KDD, and UNSW-NB15. MSCAD consistently outperforms the baseline MemStream model, especially in recall and F1-score. The proposed dynamic thresholding mechanism enhances adaptability to evolving data distributions, addressing limitations of fixed-threshold approaches. Results demonstrate MSCAD’s robustness across datasets of varying complexity, establishing it as a reliable solution for real-world anomaly detection tasks.

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

An Approach to Anomaly Detection with Dynamic Threshold Definition for Real-World Environments

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Gabriel Souza Marques,
  • Maynara Souza,
  • Flávio Arthur Oliveira Santos,
  • Cleber Zanchettin,
  • Paulo Novais

摘要

This paper introduces the Memory-based Streaming Classification for Anomaly Detection (MSCAD), which combines supervised memory updates with adaptive thresholding to enhance performance in streaming data environments. Evaluations were conducted on three publicly available datasets: KDD Cup 1999 (KDD99), NSL-KDD, and UNSW-NB15. MSCAD consistently outperforms the baseline MemStream model, especially in recall and F1-score. The proposed dynamic thresholding mechanism enhances adaptability to evolving data distributions, addressing limitations of fixed-threshold approaches. Results demonstrate MSCAD’s robustness across datasets of varying complexity, establishing it as a reliable solution for real-world anomaly detection tasks.