Scholarlink

Malware can cause abnormal events in computer systems, such as occupying or blocking access to resources like CPU and memory. Various malware detection systems utilize machine/deep learning methods as detectors to identify such threats. In this study, Kolmogorov-Arnold Network (KAN) models are developed to detect malware using system call sequences. The KAN models are trained and tested using the MalwarePlusClean2 subset of the AWSCTD dataset. Experimental results show that the KAN models achieve an accuracy (ACC) of 92.62%. The performance of the KAN models is compared with that of artificial neural networks (ANN) models, showing that KAN models outperform ANN models. Furthermore, the performance of the KAN models is evaluated in comparison with long short-term memory (LSTM), gated recurrent unit (GRU), support vector machine (SVM), and one-dimensional convolutional neural network (1D-CNN) models from the literature. The evaluation indicates KAN models generally perform better than SVM models. Additionally, it is observed that the KAN models achieve performance levels close to that of the LSTM, GRU, and 1D-CNN models.

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

Detection of Malware and Benign Samples in AWSCTD Dataset with Kolmogorov-Arnold Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Bilge Kağan Yazar,
  • Hami Satılmış,
  • Varol Şahin,
  • Sedat Akleylek

摘要

Malware can cause abnormal events in computer systems, such as occupying or blocking access to resources like CPU and memory. Various malware detection systems utilize machine/deep learning methods as detectors to identify such threats. In this study, Kolmogorov-Arnold Network (KAN) models are developed to detect malware using system call sequences. The KAN models are trained and tested using the MalwarePlusClean2 subset of the AWSCTD dataset. Experimental results show that the KAN models achieve an accuracy (ACC) of 92.62%. The performance of the KAN models is compared with that of artificial neural networks (ANN) models, showing that KAN models outperform ANN models. Furthermore, the performance of the KAN models is evaluated in comparison with long short-term memory (LSTM), gated recurrent unit (GRU), support vector machine (SVM), and one-dimensional convolutional neural network (1D-CNN) models from the literature. The evaluation indicates KAN models generally perform better than SVM models. Additionally, it is observed that the KAN models achieve performance levels close to that of the LSTM, GRU, and 1D-CNN models.