Classification of ransomware using different types of neural networks

Area of Science:

• Computer Science
• Cybersecurity
• Machine Learning

Background:

• Malware, particularly ransomware, poses a significant threat to computer and internet security by blocking user data access.
• Ransomware demands payment for data restoration, necessitating advanced detection techniques.
• Machine learning, especially neural networks, offers promising methods for malware and ransomware detection.

Purpose of the Study:

• To review different neural network types and their applications in malware and ransomware classification.
• To detail the methodology for implementing a multilayer perceptron model for ransomware detection.
• To evaluate the model's performance in binary malware detection and multi-family ransomware classification.

Main Methods:

• Literature review of neural network architectures for malware detection.
• Implementation of a multilayer perceptron (MLP) model.
• Testing the MLP model for binary classification (malware vs. goodware).
• Testing the MLP model for classifying nine ransomware families using feature vectors from prior research.

Main Results:

• The study discusses the performance and accuracy rates of the implemented multilayer perceptron model.
• Comparison of classification accuracy for binary detection and multi-family ransomware identification is presented.
• The effectiveness of the chosen neural network approach for ransomware detection is evaluated.

Conclusions:

• Neural networks, specifically the multilayer perceptron, are viable tools for detecting and classifying ransomware.
• The implemented model demonstrates potential in distinguishing between malware and goodware and categorizing diverse ransomware families.
• Further research and model refinement can enhance protection against evolving ransomware threats.