Abstract or Summary:
This paper presents a security framework that continuously authenticates smart homes users in order to make sure that only authorized ones are allowed to control their Internet of Things (IoT) devices while, at the same time, preventing them in case of performing abnormal and dangerous control actions. To do so, control commands under normal operation of both users and devices, are first implicitly traced to build a One Class Support Vector Machine (OCSVM) model as a baseline from which deviations (i.e., anomalous commands) should be detected and rejected, while normal observations (i.e., normal commands) should be considered as legitimate and allowed to be executed. Experiments conducted on our artificial datasets show the efficiency of such user behavior-based approach achieving at least 95.29% and 4.12% of True Positive (TP) and False Positive (FP) rates, respectively.