Ensuring privacy and confidentiality has become a primary task in the Internet of Things landscape. The number of connected objects which is growing exponentially in several areas, such as healthcare, transport as well as in industrial applications, means that the privacy of users is becoming more and more at risk. In this paper, we provide an overview of the problems generated in privacy and confidentiality in the field of IoT, and we show the role of machine learning algorithms in reducing these risks. We first discuss the problem of privacy, generated by the use of IoT devices, whether during data collection, storage, transmission and access control to data. We show, through case studies and real-world examples of IoT applications, the role of machine learning for preserving privacy in the world of IoT. The use of machine learning is not without costs, so we also discuss challenges and open issues in this context.