A Lightweight Mutual Privacy Preserving $k$-Means Clustering in Industrial IoT

In many industrial Internet of Things application schemes, participants, such as IoT devices, often share information with each others for analysis and categorization purpose. Through the analysis and statistical processing of the information, some necessary information is exchanged, e.g., the attributes in manufacturing IoT devices' gathered data. In this context, many devices are necessarily trusted when sharing information, which raises concerns about data privacy leakage. As the number of IoT devices in the network increases, the scale of pairwise keys increases rapidly. Furthermore, the limitation of device's computing ability makes it hard to perform centralized computing. To address these issues, we propose a lightweight $k$ -means clustering scheme that performs clustering with high accuracy, maintaining at a low key management cost, and securing the private attributes of each participant or the intermediate variables. In our proposed scheme, we use a private optimal initial cluster center generation algorithm based on attribute weights, in order to achieve the better clustering quality. Secondly, we securely find the nearest clustering center for each participant. We dynamically split and merge clustering centers respectively, in order to meet the optimal clustering result. Last but not least, we calculate the clustering centers without revealing any private attributes about the participants' information during the process. The analysis indicates that our proposed scheme can resist collusion attacks and ensures that the cost of overall pairwise key management and computation cost performed in data center are relatively low.