Attention-Based Malware Detection Model by Visualizing Latent Features Through Dynamic Residual Kernel Network.

Area of Science:

• Cybersecurity
• Computer Science
• Machine Learning

Background:

• Traditional malware analysis faces challenges in accuracy, false negatives, and resource consumption.
• Signature-based detection is often evaded by malware developers.
• Static and dynamic analysis methods have inherent limitations, including software dependency and concealment in virtual environments.

Purpose of the Study:

• To develop a novel, software-independent method for malware classification.
• To enhance the accuracy and efficiency of malware detection.
• To address the limitations of existing static and dynamic analysis techniques.

Main Methods:

• Malware files are transformed into image representations.
• A classifier is trained using neural network methodologies on these malware images.
• Involution technique and deep residual blocks are employed for feature extraction.

Main Results:

• The proposed image-based malware classification achieved 99.5% accuracy.
• This represents an absolute improvement of 95.65% over the equal probability benchmark.
• The method demonstrated increased effectiveness and software independence.

Conclusions:

• The image-based malware analysis approach is highly accurate and efficient.
• This novel method overcomes the drawbacks of traditional malware detection techniques.
• The technique offers a robust solution for classifying malware variants.