Answers are grounded in this paper and cite their sources.
A novel method enhances feature selection against adversarial attacks, achieving 98.6% accuracy in Android malware detection.
The escalating threat of Android malware has led to the widespread adoption of machine learning (ML) techniques for detection. While ML-based methods have demonstrated strong performance, they remain vulnerable to adversarial attacks designed to manipulate application features and evade detection. To address this challenge, we propose a robust and adaptive feature selection mechanism based on a Genetic Algorithm enhanced with Rank-Based Adaptive Mutation (GA-RAM). In this method, the mutation rate is dynamically adjusted according to the performance of feature subsets: lower-ranked subsets undergo more aggressive mutations to eliminate features that are likely to be injected by adversaries and are behaviorally irrelevant, while higher-performing subsets are preserved to retain important, semantically meaningful features. This adaptive balance between exploration and exploitation enables the model to converge on resilient and interpretable feature combinations. The proposed mechanism is evaluated under a comprehensive set of adversarial scenarios, including white-box (FGSM, JSMA), grey-box (mimicry, salt-and-pepper noise), black-box (GAN-based), and zero-day attacks. The system achieved 98.6% accuracy in detecting general malware and maintained over 90% accuracy across all adversarial attacks, including 94.1% for zero-day malware. Additionally, SHapley Additive exPlanations (SHAP) are employed to enhance the interpretability and trustworthiness of the detection process.
No takes yet. Share an insight, caveat, or question.
John et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: