Author(s)

Sirui Yi, Pengfei Gao, Xiang Wang and Zhiang Hu

Corresponding Author

Sirui Yi

Abstract

The rapid development of Internet of Things (IoT) not only brings lots of benefits and convenience to people, but also poses many problems and threats to people’s daily life. Among many challenges dealing with those problems, the most significant one is that it is difficult for people and even organizations to identify the types of devices connected to their internet. Under such circumstances, this research was aimed at using several machine learning (ML) algorithms, including Random forests, K-nearest neighbors (KNN), and Adaboosting, to address this challenge. This research mainly concentrated on the 46 kinds of pcap files of different devices in 1GB dataset provided by Professor Nick Feamster from the University of Chicago. To begin with, each pcap file was read as dataframe, then labeled and finally saved as csv files. Then all the data were divided into two categories: trainset (80%) and test set (20%). Finally, the method of grid research was used to select the best combination of features and parameters of each ML algorithm to achieve the best performance. Overall, the trainset and test set accuracy of our IoT device identification model is up to 99.9% and 99.2%, respectively.

Keywords

Internet of Things; Machine Learning; Smart Home; Device Identification