Authors:
Md. Rahman
1
;
2
;
Md. Delwar Hossain
1
;
Hideya Ochiai
3
;
Youki Kadobayashi
1
;
Tanjim Sakib
2
and
Syed Ramadan
2
Affiliations:
1
Division of Information Science, Nara Institute of Science and Technology, Nara, Japan
;
2
Military Institute of Science and Technology, Dhaka, Bangladesh
;
3
Graduate School of Information Science, The University of Tokyo, Tokyo, Japan
Keyword(s):
Malware Classification, Deep Learning, Vision Transformer, DCGAN.
Abstract:
Preventing malware attacks is crucial, as they can lead to financial losses, privacy breaches, system downtime, and reputational damage. Various machine learning and deep learning techniques have been proposed for malware classification. However, to evade detection, files from the same family are often altered by malware developers using various approaches so that they appear to be separate files. They may even appear as previously unidentified, commonly referred to as zero-day threats. These attacks can compromise the robustness of deep learning models trained for malware classification. In this research, we developed six fine-tuned Deep Neural Network (DNN) classifiers for classifying malware represented as images. A hybrid data augmentation technique based on Deep Convolutional Generative Adversarial Network (DCGAN) and traditional image transformation methods has been proposed to train the classifiers, enabling them to better handle malware vari-ants. A subset of the publicly ava
ilable Malimg dataset, comprising six-class and the whole dataset, were used in the experiment. Additionally, both datasets were expanded using the proposed augmentation technique to train the developed classifiers. Experimental results reveal that vision transformer-based classifiers, trained with the proposed data augmentation technique, achieve a maximum accuracy of 99.94% for six-class classification and 99.79% for 25-class classification.
(More)