A recent survey of image-based malware classification using convolution neural network

Abstract

Despite numerous breakthroughs in creating and applying new and current approaches to malware detection and classification, the number of malware attacks on computer systems and networks is increasing. Malware authors are continually changing their operations and activities with tools or methodologies, making it tough to categorize and detect malware. Malware detection methods such as static or dynamic detection, although useful, have had challenges detecting zero-day malware and polymorphic malware. Even though machine learning techniques have been applied in this area, deep neural network models using image visualization have proven to be very effective in malware detection and classification, presenting better accuracy results. Hence, this article intends to conduct a survey showing recent works by researchers and their techniques used for malware detection and classification using convolutional neural network (CNN) models highlighting strengths, and identifying areas of potential limitations such as size of datasets and features extraction. Furthermore, a review of relevant research publications on the subject is offered, which also highlights the limitations of models and dataset availability, along with a full tabular comparison of their accuracy in malware detection and classification. Consequently, this review study will contribute to the advancement and serve as a basis for future research in the field of developing CNN models for malware detection and classification.