Evaluation of Accuracy and Security on Deep Learning Model by Utilizing K-Anonymization With Attribute Selection for Reducing Computational Cost
Rei Ueda, Kota Yoshida, Takeshi Fujino · 2024
Personal data are used for various purposes but contain private information, making it difficult for the data to be collected and shared. Anonymization is a technique for securely sharing such private information, and k-anonymization is a typical method. Although k-anonymization ensures data privacy, its computational cost is high. In this paper, we apply attribute selection as a pre-processing to reduce the computational cost of k-anonymization. Deletion of some attributes of private data in advance reduces the number of combinations to be considered in the k-anonymization process. We show that applying attribute selection to k-anonymization reduces the computational cost of anonymization without decreasing the data’s utility or anonymity.