Tackling the Emerging Threat of Image-Based Steganographic Malware with Advanced Machine Learning Techniques
Savita Adhav, Ashish Sarpate, Samrat Gaisamudre, Sakshi Tirmanwar, Smitanshu Ukey · 2024
The increasing threat of malware concealed within images through steganography underscores the need for advanced detection methods. Traditional signature-based techniques often fall short as malware evolves rapidly, employing increasingly sophisticated hiding techniques. This paper investigates the application of machine learning (ML) for detecting steganographic malware in images, capitalizing on ML’s capability to discern complex patterns and adapt to emerging threats. The study explores various ML-driven approaches for image- based malware detection, including subspace-based methods for image pattern classification, ML classifiers targeting malicious JPEGs, and Attention Capsule Networks for detecting scams. Additionally, the paper reviews advancements in enhancing the robustness of AI-based malware detection through adversarial machine learning and evaluates hybrid deep learning methods for more effective malware classification. Future research directions are identified, including the development of advanced cross-validation techniques to boost model performance and generalizability, hyperparameter-tuning methods for improved optimization, and dynamic detection strategies incorporating behavioral analysis. Real-time detection with online learning is highlighted for its potential in continuous adaptation to new threats. These prospective advancements aim to address challenges related to data, optimization, real-time performance, and model interpretability, ultimately advancing the capabilities of image-based malware detection.