Enhancing Malicious Activity Classification of IoT Network Traffic Characteristics using Stacked Ensemble Learning

초록

With the expanding diversity of Internet of Things (IoT), more IoT networks are now being attacked easily by aggressors without being easily distinguished as they adapt network traffic characteristics similar to the victim's network. The consequences are the opportunities to gain control of the secure communication antagonistically, influencing internal information and prompting harms to the physical components of the IoT framework without being distinguished quickly. Anomaly-based Identification (AID) has enabled the possibility of solving this problem by helping to detect novel attacks by skimming and authenticating the IoT network designs, with higher accuracy using machine learning (ML) classification. In this work, a stacked ensemble (SE) machine learning classification approach is utilized to investigate IoT network traffic characteristics for improving novel detection of malware and noxious exercise in gadgets. The SE classifier was composed of base and meta-classifiers. The base level consisted of four classifiers; decision tree (DT), gaussian support vector machine (GSVM), ensemble DT, and naive bayes (NB). This classifiers are applied to IoT Network Intrusion Dataset from HCRLab, Korea University and sent trained data to meta classifier which was further employed to detect binary classification on test data. The experimental results shows that SE classifier achieved 99.8% accuracy with lower loss value 0.0020861 and 0.9994 precision. This application can be promising insights into the AID-based security of IoT for effective industry and/or smart factory management.