LODC: A Lightweight Online Update Method for Density-Based Clustering

Density-based clustering methods, known for their capability in detecting clusters with arbitrary geometries, consistently attract significant research attention for enhancing their computational efficiency and accuracy. Most existing research focuses on batch processing of static datasets. In recent years, real-time clustering computation has become increasingly critical in various scenarios, such as traffic planning and environmental monitoring. However, the incremental computation method for real-time processing faces the following two issues: (1) The computational accuracy cannot be guaranteed because tracking cascading cluster transformations is difficult. (2) The computational efficiency is hindered due to the uncertainty in range searches during updates, which substantially increases processing time. To overcome these issues, this paper proposes a novel Lightweight Online update method for Density-based Clustering (LODC). Specifically, LODC optimizes density-reachability and clustering-related concepts to streamline the tracking of clustering evolution. Additionally, LODC employs an evaluation model to minimize the regions involved in range searches, thereby providing a lightweight computational approach. Extensive experiments demonstrate that LODC outperforms existing methods regarding update latency and achieves robust clustering quality.