Malicious Code Variant Identification Based on Multiscale Feature Fusion CNNs.

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Malware poses a significant threat to network security due to its increasing volume and diversity.
• Deep Convolutional Neural Networks (CNNs) are effective for malware binary detection, but existing methods suffer from poor feature extraction, low accuracy, and high detection times.
• There is a need for improved malware classification techniques that can handle variants and evasive malware.

Purpose of the Study:

• To propose a novel deep learning approach, Multiscale Feature Fusion Convolutional Neural Networks (MFFCs), for effective malware classification.
• To enhance malware detection capabilities against variants and confusing malware types.
• To improve feature extraction, classification accuracy, and reduce detection time in malware analysis.

Main Methods:

• Malware code binaries are converted into grayscale images.
• The MFFC model is utilized for image normalization and feature extraction.
• Deep learning techniques are applied for identifying malware families based on visual representations.

Main Results:

• The MFFC model achieved a high accuracy of 98.72% on the Malimg dataset.
• The average detection time was significantly reduced to 5.34 milliseconds.
• Comparative experiments demonstrated superior performance over existing advanced methods.

Conclusions:

• The MFFC approach offers excellent feature extraction capabilities for malware analysis.
• The proposed method provides higher accuracy and lower detection time compared to current techniques.
• MFFCs effectively identify malware and detect variants, enhancing network security.