PrivacyGuard: a clustering-based approach for privacy-preserving IoT sensor data publishing

The Internet of Things (IoT) has become a backbone of modern life, where the integration of IoT devices is being used in many of our daily infrastructures from energy grids and water distribution systems to healthcare and household devices. The adaptation of machine learning in modelling IoT sensor data enables smarter and more efficient decision-making. As a result, developing machine learning-based models has become essential, which in turn requires more industrial data to be available for researchers and developers. However, most, if not all, organizations are reluctant to share sensor data due to concerns about exposing sensitive information or the risk that hackers or competitors may exploit their data for their own interest. In this article, we propose PrivacyGuard, a novel privacy-preserving approach that enables secure data sharing and encourages sensor data owners to publish their data without risking privacy disclosure. PrivacyGuard is based on a clustering technique for the transformation process, specifically adopting a various-width clustering (VWC) approach. This technique efficiently clusters each feature into thousands of micro-clusters while preserving data utility. PrivacyGuard replaces original values with the IDs of the generated clusters. Although these IDs are merely labels, they encode that values within the same cluster share comparable behaviour. We conduct a comprehensive evaluation on four well-known IoT datasets. Experimental results demonstrate that PrivacyGuard, compared to state-of-the-art baselines, achieves high classification accuracy and exhibits strong resistance to re-identification privacy attacks when benchmarked against the original data. Furthermore, PrivacyGuard maintains an efficient computational cost during the transformation process.