loading page

Detection of Unknown Malicious Microsoft Office Documents based on Hidden Feature Extraction by using Machine Learning
  • Kamran Saeed,
  • M. Fatih Adak
Kamran Saeed
Sakarya Universitesi

Corresponding Author:[email protected]

Author Profile
M. Fatih Adak
Sakarya Universitesi
Author Profile


The detection of unknown malicious non-programmable executable Microsoft Office files is essential for maintaining the security of computer systems and networks. Despite implementation and subsequent releases of new security protocols in Microsoft Office, documents-based viruses are still common in 2023. Most of these attacks are carried out using Microsoft Office documents. Recently, Non-Programmable Executable (NPE) in Microsoft Office documents have been used to attack many organizations. With the help of minor changes in the behavior of these office documents, document-based viruses make antivirus useless in detecting them. This paper proposes a machine learning approach, artificial intelligence-based anti-malware that can be used to detect the presence of malicious entities inside Microsoft Office documents. The detection capabilities of the anti-malware enhance over time. With the help of machine learning and hidden feature extraction (HFEM) and analysis, a malware detection model is designed to detect any malicious activity inside a Microsoft Office file. To address that issue, the model proposed in this paper integrates self-learning techniques that can be used by antivirus teams during their research while improving the detection capabilities of the antivirus software. The proposed model detects whether the files are malicious or benign and ensures that no files bypass the antivirus and harm the user. The proposed model achieved 99.9% accuracy in detecting malicious files, which is comparatively better than most existing antivirus software. The processing speed is five files per second which are helpful in terms of saving time.
24 Oct 2023Submitted to Security and Privacy
24 Oct 2023Assigned to Editor
24 Oct 2023Submission Checks Completed
24 Oct 2023Review(s) Completed, Editorial Evaluation Pending
02 Nov 2023Reviewer(s) Assigned