Domain shift severely impacts Android malware detection models

Researchers have identified significant domain shift issues in machine learning models used for Android malware detection. While models trained on one dataset (PerMalDroid) performed well on another (NATICUSdroid), the reverse showed a substantial accuracy drop. Explainable AI analysis revealed that feature importance is unstable and models rely on dataset-specific permissions, indicating fundamental mismatches in predictive feature sets. A hybrid training strategy combining common features successfully improved cross-domain performance, highlighting the need for robust malware detection systems. AI

IMPACT Highlights critical limitations in current ML-based malware detection, necessitating more robust, cross-domain adaptable systems.

RANK_REASON Research paper detailing a novel approach to diagnosing and mitigating domain shift in machine learning models for Android malware detection. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• android
• Md Rafid Islam
• NATICUSdroid
• PerMalDroid

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →