Scholarlink

With the frequent occurrence of global security incidents, the detection of malware has become a popular topic in the cyber security area. Deep learning techniques outperform conventional pattern-matching techniques in malware classification tasks. However, deep learning requires more time for model training. Therefore, we adopt transfer learning, where pretrained models are in a new model, which reduces the training time and generalization error. In this study, we propose a convolutional neural network (CNN)–support vector machine (SVM) deep learning method based on the transfer learning architecture for malware classification. VGG16, pretrained on ImageNet via transfer learning, is employed as the CNN model. We remove the fully connected layer from the VGG16 model to treat it as a feature extractor. Next, the Microsoft BIG 2015 dataset (10,868 malware samples of 9 families) is converted into grayscale images and input into the VGG16 model for feature extraction, thereby obtaining a new dataset. Finally, the new dataset is input into the SVM classifier for the classification task.

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

A Novel Malware Classification Using CNN-SVM Deep Learning Method Based on Transfer Learning Architecture

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Guan-Yu Wang,
  • Hung-Jui Ko,
  • Wei-Jen Wang

摘要

With the frequent occurrence of global security incidents, the detection of malware has become a popular topic in the cyber security area. Deep learning techniques outperform conventional pattern-matching techniques in malware classification tasks. However, deep learning requires more time for model training. Therefore, we adopt transfer learning, where pretrained models are in a new model, which reduces the training time and generalization error. In this study, we propose a convolutional neural network (CNN)–support vector machine (SVM) deep learning method based on the transfer learning architecture for malware classification. VGG16, pretrained on ImageNet via transfer learning, is employed as the CNN model. We remove the fully connected layer from the VGG16 model to treat it as a feature extractor. Next, the Microsoft BIG 2015 dataset (10,868 malware samples of 9 families) is converted into grayscale images and input into the VGG16 model for feature extraction, thereby obtaining a new dataset. Finally, the new dataset is input into the SVM classifier for the classification task.