Internet of things (iot) identifying system and associated methods

Abstract

A wireless Internet-of-Things (IoT) device identification method and framework incorporates machine learning (ML) techniques with information from the protocol used (e.g., Bluetooth, Bluetooth Low Energy/Bluetooth Smart, and others). A passive, non-intrusive feature selection technique targets IoT device captures with an ML classifier selection algorithm for the identification of IoT devices (i.e., picking the best performing ML algorithm among multiple ML algorithms available). Using an input training label and training dataset (e.g., training wireless IoT packets) associated with the IoT device, a classifier and a filter are selected. An inter-arrival-time (IAT) associated with the filtered training data set and a density distribution for the IAT are then calculated. After converting the density distribution to the training feature vector, a prediction model and the selected classifier are stored for subsequent application to testing datasets to detect and display an association of the testing dataset to a matched IoT device (i.e., identification).