Privacy Preserving BIRCH Algorithm under Differential Privacy

Current hierarchical clustering algorithms face the risk of privacy leakage during the clustering process for big dataset. While differential privacy is a relatively recent development in the field of privacy-preserving data mining, offering more robust privacy guarantees. In the paper, BIRCH algorithm under differential privacy is studied and analyzed. Firstly, Diff-BIRCH algorithm which directly add Laplace noise to the dataset before clustering is proposed. Though Diff-BIRCH algorithm achieves privacy protection, clustering result turns to be less available. Since cluster information such as cluster diameter or cluster distance may disclose during clustering process in BIRCH, three improved BIRCH algorithms under differential privacy are then designed aiming at avoiding leakage of the cluster information. Finally, experiment results validate the effectiveness and applicability of the proposed algorithms under the premise of meeting privacy budget.