Evaluation of Machine Learning Algorithms for Malware Detection.

Area of Science:

• Cybersecurity
• Machine Learning
• Software Engineering

Background:

• The increasing sophistication and volume of malware necessitate advanced detection methods.
• Manual malware analysis is becoming ineffective against rapidly evolving threats.
• Dynamic analysis and machine learning offer a promising approach to automated malware detection.

Purpose of the Study:

• To investigate the effectiveness of machine learning algorithms for dynamic malware detection.
• To evaluate various classifiers for identifying malicious software based on behavior.

Main Methods:

• Employed dynamic analysis in a simulated environment to capture malware behavior.
• Converted behavioral data into sparse vector models for machine learning.
• Utilized kNN, DT, RF, AdaBoost, SGD, extra trees, and Gaussian NB classifiers for analysis.

Main Results:

• Random Forest (RF), SGD, extra trees, and Gaussian NB classifiers achieved 100% accuracy.
• These top-performing classifiers also demonstrated perfect precision (1.00) and recall (1.00), with excellent F1-scores (1.00).

Conclusions:

• Autonomous behavior-based malware analysis combined with machine learning is highly effective.
• The proposed methods enable rapid and accurate identification of dynamic malware threats.