Unsupervised and Cost-Effective Learning: Dynamically Expose Anomaly Devices

摘要

The defect detection may be crucial issues when monitoring a large number of IoT devices, such as webcams, home appliances, and computers. The malfunctioned issues, disconnection or powerless, are obvious to detect intuitively, but slightly abnormal devices are ambiguous to define and possibly leads to critical sabotage. In the worst situation, abnormal behavior may be caused by device hijacking and authentication breaking. Our research utilizes unsupervised learning approach to distinguish those ambitious behavior with dynamical algorithm. The result shows how to quickly explore suspicious targets within large numbers of devices with cost-effectiveness computation.