Unmasking the Lurking: Malicious Behavior Detection for IoT Malware with Multi-label(…)
summary
MaGraMal identifies four fine-grained lurking behaviors from a manually annotated 5,000+ IoT malware corpus built in three person-months and trains over 224 algorithm combinations to optimize multi-label prediction. The system uses static function call graphs, computes centrality and a mask radius to drop distant nodes via graph masking, and combines structural and conventional static features. Hybrid features boost accuracy by about 5% and masking by 4.16%, overall accuracy reaches 75.83%, and runtime is roughly 12x faster than a dynamic sandbox.
AI-written summary · about
chapters (9)
- 0:00IoT malware detection intro
- 2:11Research question and setup
- 4:02Malicious behavior taxonomy
- 5:31Graph masking for call graphs
- 6:12Multi-label classification overview
- 7:19Effectiveness evaluation
- 8:44Comparison with existing methods
- 9:49Conclusion and limitations
- 11:21Q&A
related talks
48mCatch me if you can: Building a Web Malware Analyzer using Machine Learning
30mData Driven Security
27mA Machine Learning approach for detecting a Malware
46mHow to detect Malware in Encrypted Traffic Without Decryption
44mUnmasking APTs: An Automated Approach for Real-World Threat Attribution
21mMachine Learning and Cyber Security - Detecting malicious URLs in the haystack
50mDeep Impact: Recognizing Unknown Malicious Activities from Zero Knowledge
32mAI/Machine Learning: What is Actually Working in Cybersecurity?