Scholarlink

Malware continues to pose a significant threat to cybersecurity as it rapidly evolves in both structure and behavior, making early detection and accurate classification increasingly challenging. To improve the classification of malware, this study proposes a deep learning-based classification framework that leverages the EfficientNetB0 architecture for extracting discriminative features from malware images. We further integrate CycleGAN to enrich the dataset with synthetic samples. Experiments were conducted on the Malimg dataset, a publicly available benchmark from Kaggle, focusing on nine major malware families.Our model achieved an impressive accuracy of 99.41% using this approach.These results highlight the potential of combining generative and discriminative models for practical deployment in malware detection systems in the real world.

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 End-to-End Deep Learning Approach for Malware Image Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nabiha Fiza,
  • Maisha Fahmida,
  • Kaushik Deb

摘要

Malware continues to pose a significant threat to cybersecurity as it rapidly evolves in both structure and behavior, making early detection and accurate classification increasingly challenging. To improve the classification of malware, this study proposes a deep learning-based classification framework that leverages the EfficientNetB0 architecture for extracting discriminative features from malware images. We further integrate CycleGAN to enrich the dataset with synthetic samples. Experiments were conducted on the Malimg dataset, a publicly available benchmark from Kaggle, focusing on nine major malware families.Our model achieved an impressive accuracy of 99.41% using this approach.These results highlight the potential of combining generative and discriminative models for practical deployment in malware detection systems in the real world.