AI-Enhanced Differential Privacy Architecture for Securing Consumer Internet of Things (CIoT) Data

Consumer Internet of Things (CIoT) refers to the network of connected devices used by individuals in their daily lives. These devices offer benefits such as convenience, efficiency and quality of life but also create major privacy risks with the data they collect. These devices collect personal information, from daily routines to health data. Such data is a prime target for cyber-attacks as well as unauthorized access. The interconnected nature of CIoT devices amplifies the risk, as a breach in one device can compromise the entire network. Strong privacy measures such as anonymization is essential to protect users’ private data and at the same time maintaining its utility. In this context, utility refers to the preservation of data accuracy after anonymization, ensuring that the anonymized dataset remains suitable for meaningful analysis. In this article, an architecture for anonymization of data collected from CIoT devices using three-way decisions to avoid ambiguity and ensuring anonymization using differential privacy is presented. Further, two-way decisions are also explored, and their results are compared with three-way decisions, demonstrating how the latter provides a more effective solution to the privacy-utility trade-off. Finally, Stochastic Gradient Descent (SGD) is employed to determine the privacy budget $(\epsilon)$ parameter of differential privacy. The results show that using SGD to determine epsilon provided a better balance between data privacy and utility.