Enhancing Malware Detection Using Novel Feature Encoding in Machine Learning
Kanugu Sandya, G R Ashitha, Bodasu Sharath Kumar, Bussa Varun Kumar, Thatichettu Bharadwaj · 2025
Executable files like .exe, .dll, and .bat are frequently used to distribute threats like ransomware, spyware, and data-stealing programs. Attackers rely on techniques like encryption and code obfuscation to avoid detection. Traditional detection methods-particularly signature-based detection struggles to identify new threats and needs constant updates, while behaviourbased methods use more system resources and often respond only after the malware has started causing harm. Although machine learning offers better accuracy, it requires large datasets and remains vulnerable to sophisticated attacks. This work introduces a compact feature encoding method that captures the essential traits of malicious PE files. When combined with machine learning models, it improves detection accuracy while reducing false positives. Results from benchmark datasets show that the approach is more effective and reliable than traditional detection methods, offering a practical solution for identifying malware in Windows systems.