MalBuster: Scalable, Real-Time, and Concept Drift-Adaptive Malware Detection for Smart Environments
Jingwen Wang, Peilong Li, Ethan Weitkamp, Yusuke Satani, Adam Omundsen · 2024
Securing connected devices in smart environments is crucial in the age of Internet of Things (IoT). This paper proposes “MalBuster”, a scalable and real-time malware detection system created for IoT devices in smart homes, communities, and cities. MalBuster scales with increasingly large amounts of IoT data with efficiency and fault-tolerance by utilizing Apache Kafka and Apache Spark. To obtain optimal malware detection accuracy, we compare a suite of five machine learning and five deep learning models. The CNN+LSTM model achieves a superior performance of 0.992 F1 score among all. Additionally, Intel oneAPI enables faster inference at the network edge, providing over 3x speedup during deployment. For concept drift detection, the system uses a drift adaptation algorithm, which efficiently adapts to the changing malware landscape. Its adaptability is demonstrated by three simulated concept drift scenarios, which show great accuracy recovery with all scenarios obtaining above 97% accuracy.