Multi-sensor Data Privacy Protection with Adaptive Privacy Budget for IoT Systems
Xinyi Liu, Zheng Ye, Zhengxiong Li, Yidan Hu · 2024
In the era of pervasive sensing and data-driven decision-making, the Internet of Things (IoT) has become ubiquitous, with sensors serving as the fundamental building blocks of IoT devices. However, sensor readings may contain sensitive personal information or be used to infer such information, raising significant privacy concerns. Local Differential Privacy (LDP) has become the de facto standard for numerical data privacy protection. To safeguard sensor readings in IoT systems, existing LDP solutions distribute the privacy budget evenly across multiple sensors for random perturbation. Unfortunately, this approach inevitably introduces excessive noise, significantly reducing the quality of IoT services.To address this deficiency, we propose impact-aware multi-sensor data privacy protection (IMapp) to provide rigorous privacy protection for sensor readings while maintaining high-quality IoT services. IMapp leverages the fact that sensor readings from different types of sensors have varied impacts on IoT services, adaptively distributing the privacy budget across multiple sensors according to their impacts. This approach enhances the quality of IoT services while ensuring guaranteed privacy protection. Additionally, IMapp incorporates a novel LDP mechanism that ensures rigorous privacy protection for sensors with arbitrary bounded domains. Theoretical analysis and evaluation results from three collected real datasets demonstrate that IMapp achieves the same level of multi-sensor data privacy as the existing solution while improving data fusion accuracy by up to two orders of magnitude.