.png){kind=icon}
.png){kind=icon}
Return
TWStream: Three-Way Stream Clustering
.png){kind=icon}
PRE
AI
DOI：10.1109/TFUZZ.2024.3369716.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
A bunch of stream clustering algorithms have been proposed recently to mine data streams generated at high speeds from hardware platforms and software applications. Density-based methods are widely used because they can handle outliers and capture clusters of arbitrary shapes. However, it is still hard to effectively identify multidensity clusters with ambiguous boundaries in a data stream. To address these limitations, this article introduces a data stream clustering algorithm called TWStream, based on the three-way decision theory. It is a two-stage clustering algorithm based on density. In the online stage, an augmented knn graph is maintained incrementally to accelerate the update of the knn graph. In the offline stage, TWStream introduces the concept of boundary confidence to detect cluster boundaries efficiently and reveal potential cores of clusters. It integrates the skewness and sparsity of the data distribution, as well as the evolving trend of the stream. In the next step, a microcluster-based three-way clustering strategy is applied to reconstruct latent clusters. It improves the clustering quality of boundary-ambiguous clusters in a stream using a mutual reachability-based clustering approach and a three-way assignment approach. The proposed algorithm is compared with 9 competitors on 15 data streams. Experimental results show TWStream achieves competitive performance, verifying its effectiveness.
Keywords:
Data stream
density-based clustering
three-way clustering
three-way decision
uncertain data analysis
Data stream
density-based clustering
three-way clustering
three-way decision
uncertain data analysis
Journal
IF:
11.9
Papers:
5.0K
Citations:
2.9W