Speaker
Description
Personal data anonymization is an important step in dataset preprocessing, especially when dealing with sensitive information. However, the impact of this process on the quality of clustering remains poorly understood. The presented study analyzes the impact of different anonymization techniques affect the clustering results. The experimental part of the work is based on the application of ISODATA, maximin distance (Maximin) and hierarchical clustering algorithms to different datasets. The results obtained demonstrate that, for a limited number of features, depersonalization contributes to a clearer separation of the resulting clusters and while preserving the overall data structure and its trend. These findings indicate a future problem with the risks of personal data de-identification.
