Preserving privacy in wearable devices: a framework for sensitive activity data protection with utility retention

In recent years, smartphones and smartwatches have become indispensable in our daily lives. Beyond their basic functionalities, these devices enhanced by both off-the-shelf and custom-made applications offer a wide range of conveniences such as health monitoring, fitness tracking, and daily activity management. However, their growing prevalence also introduces significant privacy risks, particularly in how mobile applications acquire and manage users’ sensitive data. Embedded sensors like accelerometers, gyroscopes, and GPS continuously collect personal information, which is often automatically transmitted to the cloud for storage and analysis. While cloud-based storage and analysis provide users with convenient access, insightful visualizations, and personalized recommendations, it also raises critical concerns about data privacy. Numerous studies have highlighted the potential for such data to be exploited to accurately infer and identify users’ activities, increasing the risk of unintended exposure. To address these challenges, we propose a framework designed to balance privacy preservation and data utility. Sensitive private data, crucial for identifying individuals’ sensitive activities, is modified by selectively adding pseudo-random Gaussian noise to appropriate features before leaving personal devices. This careful application of noise in our approach ensures that privacy is preserved, reducing the accuracy of privacy-sensitive activity recognition to an average of 30%, while maintaining an average utility of 71% for less privacy-sensitive activity recognition. By targeting the right features, the framework achieves a balance that safeguards user privacy without compromising the usefulness of the data.