Ensemble Learning for Detecting Attacks and Anomalies in IoT Smart Home
Shivanjali Khare, Michael W. Totaro
Abstract
The expansion of the Internet of Things (IoT) has also given rise to an increasing number of destructive attacks that pose severe threats to exposed IoT devices. As such, IoT technologies involve a number of privacy and security challenges. Anomalies in the IoT environment often lead to unexpected disruptions or sensor failures in a machine, and may facilitate opportunities of intrusion by an attacker. Thus, it is crucial to observe the unexpected events from sensor signals. In this paper, the performance of an ensemble machine learning model is compared against the traditional learning approach for attack and anomaly detection in a smart home IoT environment on a categorical DS2oS traffic traces dataset. After training each model with the training dataset, we combine the best models using the AdaBoost ensemble learning approach in order to obtain optimized results for anomaly detection. This paper discusses various evaluation models and metrics, along with a clear description of training process. We also compare our results with other similar existing models. Finally, we discuss challenges and future work.