Deep Learning Approach to Malware Multi-class Classification Using Image Processing Techniques
Mamta Kumari, George Hsieh, Christopher A. Okonkwo · 2017
Malicious software has been growing exponentially during the past years. One of the major challenges for antimalware industry is the vast amounts of data and files which need to be evaluated for potential malicious content. To effectively analyze such large amounts of files, machine learning based malware classification approaches have been developed to classify malware into families based on the same forms of malicious behaviors. This paper presents our design and implementation of a malware classification approach using the Convolutional Neural Networks (CNNs), a prime example of deep learning algorithms. It makes use of CNNs to learn a feature hierarchy for classifying samples of malware binary files, represented as gray-scale images, to their corresponding families. It also uses transfer learning techniques to facilitate model building. Three different models of CNNs were developed and these implemented methods achieved validation accuracy around 97% using the large malware dataset provided for the Microsoft Malware Classification Challenge (BIG 2015).