Research On IoT Devices Fingerprinting Using Neural Networks

In recent years,the development and application of Internet of things technology are profoundly changing People’s Daily life.It continues to expand in areas such as smart homes,smart cities,industrial systems,connected health products,and so on.Unlike traditional Internet services that only serve user’s online activities,IoT devices can recognize or directly interact with users’ physical activities(e.g.,unlocking door)in the close range.Smart home is an important application of the Internet of things,where multiple IoT devices work together to facilitate all kinds of user activities by sensing surroundings,interpreting human commands and providing feedback.However,at the same time,this rapidly developing technology also introduces new threats to the privacy of users.Since network packets between IoT device and remote server and within IoT devices themselves can be eavesdropped,thereby revealing the privacy of users.The main research work of this paper is as follows:we look into how users’ private information can be leaked from network traffic generated in the smart home network.Although researchers have proposed techniques to infer IoT device types or user behaviors under clean experiment setup,the effectiveness of such approaches become questionable in the complex but realistic network environment,where common techniques like Network Address and Port Translation(NAPT)are enabled.Therefore,traditional methods such as the classical machine learning model are much less efficient for traffic analysis,as the features picked manually are not distinctive any more.Inspired by the continuous development of Deep learning,we apply Deep Neural Network(DNN)to Network traffic analysis and device fingerprint recognition.To this end,we designed three DNN model architectures(CNN,LSTM and CNN-LSTM),trained the parameters and evaluated the performance with three different data sets under three different network configurations,namely NAPT,NAPT+VPN and wireless network.Then,a traffic analysis system,HomeMole,is proposed to automatically infer the Internet of things devices behind the smart home network and the user activities interacting with it.the results showed that our system was able to differentiate device types and infer device activities with high accuracy in every scenario we tested.We believe the privacy concerns about smart home is worth paying attention to and new measures should be built by the involved entities,like IoT and network appliance vendors,to protect user privacy in smart home.