CNN-LSTM Based Malicious Code Detection
2
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This paper proposes a hybrid CNN-LSTM model for malicious code detection, combining static and dynamic analysis. CNN extracts spatial features from grayscale images of malware binaries, while LSTM captures temporal behavior from system call sequences. The model achieves high accuracy (94.6%) and F1-score (93.2%) on public datasets, outperforming traditional and single-stream deep learning methods. Its dual-channel design enables comprehensive feature representation, enhancing robustness against obfuscation and behavioral variation. The approach demonstrates strong potential for practical deployment in intelligent malicious code detection systems. Future work will explore attention mechanisms and graph-based modeling to improve detection precision and interpretability.
Abstract
This paper proposes a hybrid CNN-LSTM model for malicious code detection, combining static and dynamic analysis. CNN extracts spatial features from grayscale images of malware binaries, while LSTM captures temporal behavior from system call sequences. The model achieves high accuracy (94.6%) and F1-score (93.2%) on public datasets, outperforming traditional and single-stream deep learning methods. Its dual-channel design enables comprehensive feature representation, enhancing robustness against obfuscation and behavioral variation. The approach demonstrates strong potential for practical deployment in intelligent malicious code detection systems. Future work will explore attention mechanisms and graph-based modeling to improve detection precision and interpretability.