Density peak value clustering method for RDPC reconstruction density peak value

The method of reconstructing density peaks by RDPC addresses the shortcomings of density clustering algorithms in identifying complex clusters, achieving effective clustering of clusters of arbitrary shapes and clusters with large density differences, improving clustering performance and reducing algorithm complexity.

CN118940069AInactive Publication Date: 2024-11-12LANZHOU JIAOTONG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410946230.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-15
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing density clustering algorithms struggle to effectively identify inconsistencies in the density distribution of complex clusters, resulting in poor clustering performance, especially in datasets containing complex clusters where it is difficult to accurately identify cluster centers.

Method used

The density peak clustering method, which reconstructs density peaks using RDPC, reconstructs the density distribution of data points by calculating the Euclidean distance and k-nearest neighbors between data points, so that the density distribution of cluster center points exhibits the characteristics of high density at the center and low density at the edge, and the clustering is completed by label propagation.

Benefits of technology

It improves the clustering effect on clusters of arbitrary shapes, clusters with large density differences, and clusters with multiple centroids, reduces the algorithm's time complexity, and can automatically identify cluster centroids without needing to know the number of clusters in advance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118940069A_ABST
    Figure CN118940069A_ABST
Patent Text Reader

Abstract

The invention discloses a density peak value clustering method for reconstructing a density peak value through an RDPC, and provides a density peak value clustering algorithm RDPC for reconstructing the density peak value, and the algorithm unifies the density distribution of clusters in a data set through the density distribution rule of reconstructed data points. Therefore, the cluster distribution better presents peak density distribution with high density of the central region and low density of the edge region, and the clustering effect of the algorithm is better improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a density peak value clustering method for RDPC reconstruction of density peak values. Background Art

[0002] Clustering is an unsupervised learning method that can divide data into multiple clusters with high similarity within the same cluster and low similarity between different clusters, so as to mine and analyze the information of the data. In the real world, clustering is widely used in data analysis, text extraction, behavior detection and atmosphere monitoring. For the identification of complex clusters, density clustering often shows better results.

[0003] Density clustering is a method to identify complex clusters in a data set by calculating the density of data points or discovering the density distribution pattern of data points in a cluster. By using different density kernel calculation methods to increase the density of the core area and reduce the density of the edge area, clusters of arbitrary shapes can be identified, and the algorithm can further divide data sets containing clusters with large density differences.

[0004] Data sets containing complex clusters often do not have a relatively uniform density distribution, which makes the density distribution of clusters inconsistent and it is not possible to identify a uniform density law to identify different clusters. In view of this situation, changing the density distribution to identify the cluster center of each cluster can reduce the impact of the irregular density distribution, so that the algorithm can better identify complex clusters in the data set. For example, Liu et al. proposed a density kernel based on shared k-nearest neighbors, which can better identify cluster centers on data sets of different densities. Lotfi et al. introduced a score based on a fuzzy kernel to identify cluster centers. Zhang et al. defined another density kernel that can well reflect connectivity to identify cluster centers. The density normalization (DN) clustering algorithm aims to narrow the density gap between cluster centers with significant density differences.

[0005] The present invention provides a density peak clustering algorithm RDPC for reconstructing density peak values. The algorithm reconstructs the density distribution law of data points and unifies the density distribution of clusters in the data set, so that the cluster distribution better presents a peak density distribution with high density in the central area and low density in the edge area, thereby greatly improving the clustering effect of the algorithm. Summary of the invention

[0006] The purpose of the present invention is to solve the above-mentioned problem and to provide a density peak clustering method for RDPC reconstructing density peaks.

[0007] To achieve the above object, the technical solution adopted by the present invention is:

[0008] This includes clustering the Zigzag dataset using RDPC, as follows:

[0009] ① Based on the input Zigzag data set, calculate the Euclidean distance Dist between all data points in the data set n×n , where n is the number of points in the Zigzag dataset;

[0010] ②According to Dist n×n , get the k nearest neighbors kNN of all points;

[0011] ③ Calculate the density ρ of all data points according to formula (1);

[0012]

[0013] in Represents point x i k nearest neighbors, d ij Represents point x i and point x j The distance, ρ i Represents data point x i density;

[0014] ④ From the lowest density and unvisited point x i First, according to the rule of formula (2), select Inner and x i The distance is less than d r The minimum density point x j As the next point of the reconstructed density, x i The density ρ i Add to x j The density ρ j superior, Incorporation x i and x j Set to visited, set x j Becomes the new x i , Become new Continue to select points according to the rule of formula (2) until there are no points that meet the conditions. The last point that meets the conditions is recorded as a cluster center point, and all points within the kNN of the last point are marked as visited. Then continue to run ④ until all points are visited;

[0015]

[0016] in Represents point x i k nearest neighbors, d ij Represents point x i and point x j The distance, ρi Represents data point x i The density of j Represents data point x j The density, d r is the input distance threshold.

[0017] ⑤All cluster center points obtained from ④ are all cluster center points of the Zigzag data set;

[0018] ⑥The density reconstructed in ④ is the new density ρ of all points in the Zigzag dataset i ';

[0019] ⑦ Assign separate labels to all the cluster center points of Zigzag, and then traverse from high-density points to low-density points, assigning each data point the label of its nearest high-density point;

[0020] ⑧Finally, the data points with the same label are placed into a cluster to complete the clustering of the Zigzag data set.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] Compared with the current newer density-based algorithms, the RDPC algorithm of the present invention can identify clusters of arbitrary shapes, and the proposed enhanced density method can reconstruct the density distribution of clusters, so that the algorithm can better identify data sets containing clusters with large density differences and data sets containing clusters with relatively uniform density. The algorithm has low time complexity and good clustering effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a 3D density comparison diagram before and after the updated density distribution of the present invention;

[0024] Figure 2 It is a schematic diagram of clustering results of the Zigzag data set of the present invention under different algorithms;

[0025] Figure 3 It is a schematic diagram of the clustering results of the Pathbased data set of the present invention under different algorithms;

[0026] Figure 4 The clustering method flow chart of the present invention. DETAILED DESCRIPTION

[0027] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0028] according to Figure 4 As shown, in order to solve the above technical problems, the present invention provides a density peak clustering method for RDPC reconstructing density peaks.

[0029] Use RDPC to cluster the Zigzag dataset as follows:

[0030] ① Based on the input Zigzag data set, calculate the Euclidean distance Dist between all data points in the data set n×n , where n is the number of points in the Zigzag dataset;

[0031] ②According to Dist n×n , get the k nearest neighbors kNN of all points;

[0032] ③ Calculate the density ρ of all data points according to formula (1);

[0033]

[0034] in Represents point x i k nearest neighbors, d ij Represents point x i and point x j The distance, ρ i Represents data point x i density.

[0035] ④ From the lowest density and unvisited point x i First, according to the rule of formula (2), select Inner and x i The distance is less than d r The minimum density point x j As the next point of the reconstructed density, x i The density ρ i Add to x j The density ρ j superior, Incorporation x i and x j Set to visited, set x j Becomes the new x i , Become new Continue to select points according to the rule of formula (2) until there are no points that meet the conditions. The last point that meets the conditions is recorded as a cluster center point, and all points within the kNN of the last point are marked as visited. Then continue to run ④ until all points are visited;

[0036]

[0037] in Represents point x i k nearest neighbors, d ij Represents point x iand point x j The distance, ρ i Represents data point x i The density of j Represents data point x j The density, d r is the input distance threshold.

[0038] ⑤All cluster center points obtained from ④ are all cluster center points of the Zigzag data set;

[0039] ⑥The density reconstructed in ④ is the new density ρ of all points in the Zigzag dataset i ';

[0040] ⑦ Assign separate labels to all the cluster center points of Zigzag, and then traverse from high-density points to low-density points, assigning each data point the label of its nearest high-density point;

[0041] ⑧Finally, the data points with the same label are placed into a cluster to complete the clustering of the Zigzag data set.

[0042] Comparative experiment and clustering results

[0043] To verify the effectiveness of the algorithm, the algorithm was compared with 9 comparison algorithms including DBSCAN, k-Means, OPTICS, Spectral clustering (SC), DPC, MDPC, DN, Extreme and DPC-CE in clustering on 15 complex data sets and 7 UCI data sets. The 15 complex data sets are classified as follows: Jain, Mouse, Unbalance and Varydensity are data sets containing clusters with large density differences; Toy, Circles, Pathbased and Ring are data sets containing multi-center point clusters; Asymmetric, Skewed, Moon, S1 and Unbalance1 are data sets containing clusters with relatively even density; Spiral and Zigzag are data sets containing clusters of arbitrary shapes. In addition, ARI is used

[18] and NMI

[19] Evaluation indicators are used to quantitatively compare clustering results.

[0044] The experimental results are shown in Tables 1 to 4. We can observe from Tables 1 and 2 that RDPC performs better than other comparison algorithms in the four types of complex data sets, and can also better identify each cluster in the data set such as Pathbased, Toy and Zigzag, which contain multiple types of clusters.

[0045] Table 1 Quantitative clustering results of complex data sets ARI

[0046]

[0047] As shown in Tables 3 and 4, RDPC achieved the highest evaluation index in all six UCI datasets and the second highest evaluation index in the remaining Seeds dataset.

[0048] Experimental results show that RDPC has good clustering effect and can handle different types of complex data sets.

[0049] Table 2 Quantitative clustering results of complex data sets NMI

[0050]

[0051]

[0052] Table 3 Quantitative clustering results of UCI dataset ARI

[0053]

[0054] Table 4 Quantitative clustering results NMI of UCI dataset

[0055]

[0056] Figure 1 Update the density distribution for the Zigzag dataset before and after. Figure 1 (a) It can be found that after the RDPC update density distribution week, the density of the "Z"-shaped cluster in Zigzag is Figure 1 (b) The density of the cluster without density update can better show the characteristics of high density in the cluster center area and low density in the cluster edge area, so that the label of cluster center point 282 can be smoothly transferred to the entire "Z"-shaped cluster.

[0057] Figure 2 and Figure 3 Here are two examples of complex data sets. Figure 2 The clustering results of the Zigzag dataset under different algorithms. Figure 3 The clustering results of the Pathbased dataset under different algorithms are shown in Figure 2. It can be seen that for these two complex datasets, RDPC can identify clusters in the dataset and the clustering effect is good.

[0058] The present invention mainly proposes the RDPC clustering algorithm. Based on the original data point density distribution, the algorithm reconstructs the density of the data points according to certain rules, thereby updating the density distribution of the cluster, so that the label of the cluster center point can be smoothly propagated to each point in the cluster. According to the above algorithm analysis and experimental results evaluation, the beneficial effects of the technical solution of the present invention are mainly manifested in the following aspects:

[0059] ①The proposed algorithm can identify complex data sets containing clusters of arbitrary shapes, clusters with large density differences, and clusters with multiple center points.

[0060] ② The algorithm searches for cluster centers automatically, and does not require knowing the number of clusters in advance.

[0061] ③For complex data sets and UCI data sets, RDPC can obtain better clustering results.

[0062] ④The time complexity of the algorithm is, which has a low time complexity.

[0063] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

[0064] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A density peak clustering method for RDPC reconstruction of density peaks, comprising clustering a Zigzag data set using RDPC, characterized in that: The details are as follows: ① Based on the input Zigzag data set, calculate the Euclidean distance Dist between all data points in the data set n×n , where n is the number of points in the Zigzag data set; ②According to Dist n×n , get the k nearest neighbors kNN of all points; ③ Calculate all data points x according to formula (1) i The density ρ i ; in Represents point x i k nearest neighbors, d ij Represents point x i and point x j The distance, ρ i Represents data point x i density; ④ From the lowest density and unvisited point x i First, according to the rule of formula (2), select Inner and x i The distance is less than d r The minimum density point x j As the next point of the reconstructed density, x i The density ρ i Add to x j The density ρ j superior, Incorporation x i and x j Set to visited, set x j Becomes the new x i , Become new Continue to select points according to the rule of formula (2) until there are no points that meet the conditions. The last point that meets the conditions is recorded as a cluster center point, and all points in the kNN of the last point are marked as visited. Then continue to run ④ until all points are visited. in Represents point x i k nearest neighbors, d ij Represents point x i and point x j The distance, ρ i Represents data point x i The density of j Represents data point x j The density, d r is the input distance threshold; ⑤All cluster center points obtained from ④ are all cluster center points of the Zigzag data set; ⑥The density reconstructed in ④ is the new density ρ of all points in the Zigzag dataset i '; ⑦ Assign separate labels to all the cluster center points of Zigzag, and then traverse from high-density points to low-density points, assigning each data point the label of its nearest high-density point; ⑧Finally, the data points with the same label are placed into a cluster to complete the clustering of the Zigzag data set.