Adversarial Vulnerability Under Temporal Concept Drift: A Longitudinal Study of Android Malware Detection

1. 1Extract static and dynamic features from Android applications using emulator and real-device executions
2. 2Organize the dataset into yearly slices to evaluate temporal concept drift
3. 3Evaluate models under different deployment protocols, including same-year training and testing
4. 4Apply drift-aware evaluation metrics to assess adversarial robustness
5. 5Update models periodically to adapt to changing malware patterns

Security researchers and AI engineers on a team benefit from understanding adversarial vulnerability to improve malware detection models, while data scientists can apply these insights to other domains with concept drift

💡 Temporal concept drift significantly impacts adversarial robustness in Android malware detection, requiring periodic model updates

🚨 Android malware detection under attack! 🤖 Learn how to evaluate adversarial robustness under temporal concept drift 💡