The Road Less Traveled: Investigating Robustness and Explainability in CNN Malware Detection

2025-03-03

AI Summary

Convolutional neural networks (CNNs) for binary-image-based malware detection suffer from a threefold dilemma: high accuracy yet low interpretability and poor robustness. Method: This study systematically integrates three explainable AI (XAI) techniques—occlusion maps, HiResCAM, and SHAP—to enable the first multi-granularity, quantitative attribution analysis of CNN decision-making. It identifies that common obfuscation techniques can cause classification accuracy to drop by up to 50%, and accordingly proposes a feature-space perturbation–aware robustness enhancement strategy. Furthermore, it develops a heatmap-based artifact identification method leveraging anomalous saliency patterns, enabling reproducible human-in-the-loop reverse analysis. Contribution/Results: Experiments demonstrate significant improvements in model robustness against obfuscation. The approach provides security analysts with verifiable, traceable decision evidence chains—bridging the gap between automated detection and actionable forensic interpretation.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding

Abstract

Machine learning has become a key tool in cybersecurity, improving both attack strategies and defense mechanisms. Deep learning models, particularly Convolutional Neural Networks (CNNs), have demonstrated high accuracy in detecting malware images generated from binary data. However, the decision-making process of these black-box models remains difficult to interpret. This study addresses this challenge by integrating quantitative analysis with explainability tools such as Occlusion Maps, HiResCAM, and SHAP to better understand CNN behavior in malware classification. We further demonstrate that obfuscation techniques can reduce model accuracy by up to 50%, and propose a mitigation strategy to enhance robustness. Additionally, we analyze heatmaps from multiple tests and outline a methodology for identification of artifacts, aiding researchers in conducting detailed manual investigations. This work contributes to improving the interpretability and resilience of deep learning-based intrusion detection systems

Problem

• Enhancing CNN interpretability in malware detection using explainability tools.
• Addressing robustness issues caused by obfuscation techniques in CNN models.
• Proposing mitigation strategies to improve CNN resilience in cybersecurity.

Innovation

• Integrates explainability tools for CNN malware detection
• Proposes mitigation strategy against obfuscation techniques
• Analyzes heatmaps to identify artifacts in malware classification