Dynamic Malware Classification and Signature Generation Using Multi-View Convolutional Neural Networks

The increasing sophistication of malware has posed significant challenges to cybersecurity, necessitating advanced approaches for malware detection and signature generation. This study presents a novel method for dynamic malware classification and signature generation using Multi-View Convolutional Neural Networks (MVCNNs). The proposed framework captures multiple perspectives of malware behavior, including system calls, memory access patterns, and network activities, to create a comprehensive representation of malicious activities. The MVCNN architecture leverages these diverse views to enhance classification accuracy, addressing the limitations of single-view analysis. The dynamic execution of malware in a sandboxed environment generates behavior logs, which are preprocessed and transformed into multi-dimensional feature maps for input into the MVCNN. The model then extracts high-level features across multiple views, enabling robust classification and the automated generation of unique malware signatures. Experimental results demonstrate the proposed method's effectiveness, achieving a classification accuracy of over 98% on benchmark datasets. Additionally, the generated signatures exhibit high precision and recall in identifying malware variants. The proposed approach not only improves malware detection rates but also provides scalability and adaptability for real-time cybersecurity applications. This research highlights the potential of multi-view deep learning frameworks to advance dynamic malware analysis and enhance cybersecurity resilience.