An Anonymization Method to Improve Data Utility for Classification

k-anonymity is a popular method to preserve privacy in microdata, which sacrifices data utility for preserving individuals’ privacy. Therefore, how to preserve privacy with high data utility has been becoming a hot topic in k-anonymity area. Existing anonymization methods seldomly consider the data utility for specific data mining. To address the problem, we define a novel attribute weight measurement for determining the generalization order, and further propose a new anonymization algorithm based on the weight measurement using global generalization, called Weighted Full-Domain Anonymization (WFDA) Algorithm. The main idea of the algorithm is to generalize attributes with large weights to lower levels, and attributes with small weights to high levels. The proposed algorithm can reserve data utility for classification to a large extent. Experiments show that anonymous data resulted from the proposed method retains higher utility, i.e., has better classification accuracy, than that generated by other anonymization methods.