LLM-MalDetect: A Large Language Model-Based Method for Android Malware Detection
Ruirui Feng, Hui Chen, Shuo Wang, Md Monjurul Karim, Qingshan Jiang
Abstract
Android malware poses a significant cybersecurity threat, enabling unauthorized data access, financial fraud, and device compromise. Although deep learning methods are widely used for malware detection, they often struggle with stability and adaptability in the face of evolving threats. While large language models (LLMs) have shown promise in this area, their application to Android malware detection remains underexplored, particularly with regard to optimizing the semantic relationships within Android application packages (APKs). To address this gap, we introduce LLM-MalDetect, a novel framework that improves LLM-based APK analysis by explicitly modeling semantic dependencies and leveraging structured prompt engineering for optimized detection. Our approach formalizes LLM adaptation through a robust string-based feature extraction method and a tailored fine-tuning strategy to enhance precision. Evaluations on benchmark datasets demonstrate that LLM-MalDetect achieves up to 98.97% accuracy, outperforms existing methods in terms of robustness, and enables real-time analysis.