Scalable malware identification and classification using deep neural network

Abstract

Malware presents a challenge to organizations as they threaten smooth functioning of both physical and virtual system. Timely identification of malware is critical as it allows organizations to eliminate the threat before damage has been done. This paper proposes a scalable deep learning framework for classifying portable executable files as benign or malicious using file header information. The proposed method relies on the representational power of deep neural networks which allows them to learn complex characteristics found in the file header information. A deep neural network is trained using header information extracted from sample of benign and malicious files binaries. An accuracy of 0.98 and false positive rate of 0.019 were obtained.