Scholarlink

Traditional network anomaly detection methods struggle in dynamic IoT environments. We propose a novel approach for transforming time series into images using the Gramian Angular Field (GAF) for classification with a Vision Transformer (ViT). On a public IoT dataset, our model achieved 97.72% accuracy, 97.71% F1-score, and exceptionally high recall (99.34%), identifying nearly all attacks. However, this strength resulted in a higher false-alarm rate (3.85%) than that of the benchmark models. This approach is promising for prioritizing attack detection and requires further false alert reduction.

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

Network Anomaly Detection Using Gramian Angular Field Transformation and Vision Transformer

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jaroslaw Kobiela

摘要

Traditional network anomaly detection methods struggle in dynamic IoT environments. We propose a novel approach for transforming time series into images using the Gramian Angular Field (GAF) for classification with a Vision Transformer (ViT). On a public IoT dataset, our model achieved 97.72% accuracy, 97.71% F1-score, and exceptionally high recall (99.34%), identifying nearly all attacks. However, this strength resulted in a higher false-alarm rate (3.85%) than that of the benchmark models. This approach is promising for prioritizing attack detection and requires further false alert reduction.