LGAIDSMar 31, 2024

Block-Diagonal Guided DBSCAN Clustering

arXiv:2404.01341v222 citationsh-index: 5IEEE Trans Knowl Data Eng
AI Analysis

This is an incremental improvement for database mining and cluster analysis, addressing specific bottlenecks in DBSCAN.

The paper tackles DBSCAN's limitations in handling high-dimensional large-scale data by introducing an improved version that uses the block-diagonal property of similarity graphs to guide clustering, achieving superior performance on twelve benchmark datasets.

Cluster analysis plays a crucial role in database mining, and one of the most widely used algorithms in this field is DBSCAN. However, DBSCAN has several limitations, such as difficulty in handling high-dimensional large-scale data, sensitivity to input parameters, and lack of robustness in producing clustering results. This paper introduces an improved version of DBSCAN that leverages the block-diagonal property of the similarity graph to guide the clustering procedure of DBSCAN. The key idea is to construct a graph that measures the similarity between high-dimensional large-scale data points and has the potential to be transformed into a block-diagonal form through an unknown permutation, followed by a cluster-ordering procedure to generate the desired permutation. The clustering structure can be easily determined by identifying the diagonal blocks in the permuted graph. We propose a gradient descent-based method to solve the proposed problem. Additionally, we develop a DBSCAN-based points traversal algorithm that identifies clusters with high densities in the graph and generates an augmented ordering of clusters. The block-diagonal structure of the graph is then achieved through permutation based on the traversal order, providing a flexible foundation for both automatic and interactive cluster analysis. We introduce a split-and-refine algorithm to automatically search for all diagonal blocks in the permuted graph with theoretically optimal guarantees under specific cases. We extensively evaluate our proposed approach on twelve challenging real-world benchmark clustering datasets and demonstrate its superior performance compared to the state-of-the-art clustering method on every dataset.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes