Talk · International Conference on Electronics, Computer and Computation · April 10, 2026

AI-Based Detection of Evasive Malware Using Machine Learning and Deep Learning Techniques

Mohammad Imtiyaz Gulbarga, Ruslan Isaev, Ruslan Amanov

Abstract

Modern malware increasingly employs sophisticated evasion techniques, including obfuscation, polymorphism, and anti-analysis mechanisms, to bypass traditional detection systems. Memory analysis captures runtime behavioral characteristics that persist regardless of code-level alterations, limiting practical deployment. To address this crucial gap, we propose a framework that combines memory forensics with ensemble machine learning. We extended the Advanced-VolMemLyzer tool from 75 to over 250 behavioral features, targeting evasive malware indicators. A balanced dataset of 58,596 memory samples covering three malware families Ransomware, Spyware, and Trojan Horse was assembled from real-world specimens. The two-layer stacked ensemble reached 99% detection accuracy at 0.23 seconds per sample, outperforming traditional signature-based approaches (78.4%) and remaining feasible for real-time deployment.

Advanced Malware Detection TechniquesDigital and Cyber ForensicsSecurity and Verification in Computingmemory forensicsmalware obfuscationpolymorphic malware

AI-Based Detection of Evasive Malware Using Machine Learning and Deep Learning Techniques

Abstract

More by Ruslan Isaev