Differential Privacy-Preserving Density Peaks Clustering Based on Shared Near Neighbors Similarity

Abstract

Density peaks clustering is a novel and efficient density-based clustering algorithm. However, the problem of the sensitive information leakage and the associated security risk with the applications of clustering methods is rarely considered. To address the problem, we proposed differential privacypreserving density peaks' clustering based on the shared near neighbors similarity method in this paper. First, the Euclidean distance and the shared near neighbors similarity were combined to define the local density of a sample, and the Laplace noise was added to the local density and the shortest distance to protect privacy. Second, the process of cluster center selection was optimized to select the initial cluster centers based on the neighborhood information. Finally, each sample was assigned to the cluster as its nearest neighbor with higher local density. The experimental results on both the UCI and synthetic datasets show that compared with other algorithms, our method more effectively protects the data privacy and improves the quality of the clustering results.INDEX TERMS Privacy preservation, differential privacy, density peaks clustering algorithm, shared near neighbors similarity.

Effective Density Peaks Clustering Algorithm Based on the Layered K-Nearest Neighbors and Subcluster Merging

“…The parameter settings of the contrast algorithms are taken from the original articles. Cluster evaluation indexes include the ACC [33], ARI [28] and AMI [34], and the respective calculation formulas are shown below: Wine 178 13 3 Iris 150 4 3 Seeds 210 7 3 Ionosphere 351 34 2 Wdbc 569 30 2 Segmentation 2310 19 Glass 214 9 6 Libras Movement 360 91 15 Dermatology 366 34 6 Waveform 5000 21 3 Parkinsons 195 23 2 Pima 768 8 2 SCADI 70 206 7 Letter 20000 16 26 Before the experiment, to avoid fluctuations caused by inconsistent attribute values, all datasets were standardized using formula (18).…”

Sparse representation and modified density peak clustering-based state identification for multimode processes

“…There are six operation modes illustrated in table 4, where G and H are two products in the TE process, and 'G/H' denotes the quality ratio. The TE process involves 53 variables including 12 manipulated variables, 22 continuous process variables and D feed temperature Random IDV (10) C feed temperature Random IDV (11) Reactor cooling water inlet temperature Random IDV (12) Condenser cooling water inlet temperature Random IDV (13) Reaction kinetics Slow drift IDV (14) Reactor cooling water valve Sticking IDV (15) Condenser cooling water valve Sticking IDV (16) Unknown Unknown IDV (17) Unknown Unknown IDV (18) Unknown Unknown IDV (19) Unknown Unknown IDV (20) Unknown Unknown 19 composition variables. In addition, as listed in table 5, 20 faults are introduced by Ricker [32], in which 15 faults are known and 5 are unknown.…”

“…Liu et al [16] adopted logistic regression and several thresholds to detect cluster centers, in which center candidates were divided into several categories and further explored to rule out non-centers. Sun et al [17] sorted all data by an indicator and selected several of the highest data points that do not share neighbors with one another as cluster centers. However, these methods still determine the number of clusters and cluster centers through decision graphs and experience.…”

A New Density Peak Clustering Algorithm With Adaptive Clustering Center Based on Differential Privacy

“…In order to solve the problem that noisy parameters may lead to deviation between the new center point and the correct center point, reachable center point is introduced to DP-RCCFSFDP [31]. Sun et al proposed DP-DPCSNNS based on shared nearest neighbor similarity [32], which used shared nearest neighbor similarity to calculate local density and detect cluster centers with neighborhood information, thus improving the accuracy of cluster center selection.…”