A Machine Learning-Based Exploration of Relationship Between Security Vulnerabilities of IoT Devices and Manufacturers

Abstract

The internet of things has brought in innovations in the daily lives of users. The enthusiasm and openness of consumers have fuelled the manufacturers to dish out new devices with more features and better aesthetics. In an attempt to keep up with the competition, the manufacturers are not paying enough attention to cyber security of these smart devices. The gravity of security vulnerabilities is further aggravated due to their connected nature. As a result, a compromised device would not only stop providing the intended service but could also act as a host for malware introduced by an attacker. This study has focused on 10 manufacturers, namely Fitbit, D-Link, Edimax, Ednet, Homematic, Smarter, Osram, Belkin Wemo, Philips Hue, and Withings. The authors studied the security issues which have been raised in the past and the communication protocols used by devices made by these brands. It was found that while security vulnerabilities could be introduced due to lack of attention to details while designing an IoT device, they could also get introduced by the protocol stack and inadequate system configuration. Researchers have iterated that protocols like TCP, UDP, and mDNS have inherent security shortcomings and manufacturers need to be mindful of the fact. Furthermore, if protocols like EAPOL or Zigbee have been used, then the device developers need to be aware of safeguarding the keys and other authentication mechanisms. The authors also analysed the packets captured during setup of 23 devices by the above-mentioned manufacturers. The analysis gave insight into the underlying protocol stack preferred by the manufacturers. In addition, they also used count vectorizer to tokenize the protocols used during device setup and use them to model a multinomial classifier to identify the manufacturers. The intent of this experiment was to determine if a manufacturer could be identified based on the tokenized protocols. The modelled classifier could then be used to drive an algorithm to checklist against possible security vulnerabilities, which are characteristic of the protocols and the manufacturer history. Such an automated system will be instrumental in regular diagnostics of a smart system. The authors then wrapped up this report by suggesting some measures a user can take to protect their local networks and connected devices.