Efficient streaming data clustering guided by self-refined organizing map

Streaming data clustering is a significant area of research in data mining and machine learning. Unlike static data, streaming data are typically processed in chunks and more prone to dynamic cluster imbalance issues. This means that the degree of imbalance among clusters can change across different data chunks, potentially compromising the accuracy or efficiency of streaming data analysis when using current clustering methods. To address this challenge, Self-refined Organizing Map-guided Clustering (SOMAC), an effective solution for clustering imbalanced streaming data, has been introduced. SOMAC involves developing an enhanced Self-Organizing Map (SOM) to represent the overall data distribution. Initially, the SOM is refined to guide the partitioning of the dataset into numerous micro-clusters, which helps in capturing small clusters that might otherwise be missed in imbalanced data. These micro-clusters are then efficiently merged through quick retrieval using the SOM, automatically determining the true number of imbalanced clusters. Compared to existing methods for clustering imbalanced data, SOMAC is with a time complexity as low as [Formula: see text] while maintaining a highly competitive clustering accuracy. Additionally, SOMAC is both interpretable and robust to changes in hyperparameters. Extensive experiments have demonstrated its effectiveness and efficiency.