An Attribute Similarity Based Feature Vector Training for Malware Analysis and Detection in Cloud Environments

Abstract

Cloud computing is not only efficient, scalable, and flexible, but it also offers a high level of reliability on elastic resources. The IT industry makes extensive use of the platform to underpin IT infrastructure and services. One of the biggest security concerns, however, is malware attacks; certain antivirus scanners aren't able to pick up on metamorphic or encrypted malware because of the environment's complexity and scale, thus these threats can get through them. Another wave with malware attacks, this time encompassing intelligent embedded devices, has arrived with the rise of the Internet of Things (IoT). Running a full malware scanner on these devices is difficult due to the low energy resources. There is a pressing need for innovative methods of scanning mobile devices for malicious software. One service that can be offered as a cloud-based option is malware detection. Dynamic behavior-based strategies have been analyzed to facilitate the analysis of possibly dangerous software. Because the executions settings in which these tactics are carried out are artificial and do not faithfully mirror the contexts of end-users, the intended recipients of the malicious activity, they sometimes produce partial results. A novel approach using Attribute Similarity based Feature Vector Training (ASbFVT) model is proposed to facilitate more accurate behavior based analysis of potentially malicious software using the extracted and selected features in the cloud environment. This platform lets users outsource tasks like program execution and analysis to remote environments like cloud servers or security laboratories, while still maintaining control over how their nodes behaves. The evaluation showed that the proposed framework enables security labs to increase the thoroughness of the analysis through carrying out a fine-grained assessment of the behavior of the program without incurring any computational cost to end-users. The proposed model when contrasted with the traditional models performs better in feature vector generation.