Abstract
In this paper, we aimed to build the Abnormal Behavior Profiling (ABP) of IoT devices to supplement other studies that focused on the abnormal behavior detection of IoT devices with a high accuracy rate by a wide variety of machine learning algorithms. ABP will integrate all types of abnormal behavior detection and possess a key role for the purpose of IoT security in the future. Our technical motivation was derived from IoT smart sensors, which are equipped with high computing power and communication capability; moreover, it is possible that one sensed data can be modified instead of all sensed data (e.g., a sensor reporting temperature, humidity, light and voltage) for malicious purposes. In this kind of threat, it affects the detection accuracy of abnormal behavior and the degradation of the abnormal behavior detection from both machine learning algorithms, such as the k-Means algorithm and support vector machine (SVM), was observed. The k-Means algorithm and SVM were used to detect one sensed data modification out of 4 possible data points from one sensor and the results demonstrated that the k-Means algorithm (92%) had less affection than the SVM (69.5%) in terms of detection accuracy. The ABP was constructed and proposed on the basis of a k-Means algorithm with 10 clusters. In future work, we will investigate how to improve the detection accuracy of abnormal behavior in an IoT environment so that an improved ABP will be available.