On the use of artificial intelligence to deal with privacy in IoT systems: A systematic literature review

Abstract

The Internet of Things (IoT) refers to a network of Internet-enabled devices that can make different operations, like sensing, communicating, and reacting to changes arising in the surrounding environment. Nowadays, the number of IoT devices is already higher than the world population. These devices operate by exchanging data between them, sometimes through an intermediate cloud infrastructure, and may be used to enable a wide variety of novel services that can potentially improve the quality of life of billions of people. Nonetheless, all that glitters is not gold: the increasing adoption of IoT comes with several privacy concerns due to the lack or loss of control over the sensitive data exchanged by these devices. This represents a key challenge for software engineering researchers attempting to address those privacy concerns by proposing (semi-)automated solutions to identify sources of privacy leaks. In this respect, a notable trend is represented by the adoption of smart solutions, that is, the definition of techniques based on artificial intelligence (AI) algorithms. This paper proposes a systematic literature review of the research in smart detection of privacy concerns in IoT devices. Following well-established guidelines, we identify 152 primary studies that we analyze under three main perspectives: (1) What are the privacy concerns addressed with AI-enabled techniques; (2) What are the algorithms employed and how they have been configured/validated; and (3) Which are the domains targeted by these techniques. The key results of the study identified six main tasks targeted through the use of artificial intelligence, like Malware Detection or Network Analysis. Support Vector Machine is the technique most frequently used in literature, however in many cases researchers do not explicitly indicate the domain where to use artificial intelligence algorithms. We conclude the paper by distilling several lessons learned and implications for software engineering researchers.