Android malware detection by using graph optimization of static features based on pre-trained language models

doi:10.1016/j.infsof.2026.108012

Android malware detection by using graph optimization of static features based on pre-trained language models

2026-01-06

0

PRE

AI

DOI：10.1016/j.infsof.2026.108012

Original

Original request for help

Save

Abstract

En 中文

• This study proposed APSDroid, a novel method for detecting and classifying Android malware by integrating permissions-intents (PIs) and API call graphs (ACGs) with pre-trained language models (PLMs). • Forensic analysis was conducted by extracting and analyzing critical PIs and ACGs to effectively uncover normal and malicious behaviors. • APSDroid incorporates graph optimization techniques such as community detection and centrality measures to reduce graph complexity while maintaining the contextual flow of application behavior. • The approach addresses challenges in processing large-scale graph data for PLMs, including token length constraints and computational demands. • Experiments conducted on the CICMalDroid2020 dataset demonstrated the effectiveness of APSDroid, achieving an accuracy of 97.40% for malware detection and 94.23% for malware category classification. It also maintains strong resilience under obfuscation techniques, with F1-scores of 98.69% (binary) and 83.98% (multi-class), outperforming several SOTA approaches.

Android malware detection by using graph optimization of static features based on pre-trained language models

Abstract

Journal

Organization