A Weighted Granular-Ball Clustering Algorithm Based on Regularization and Entropy

In high-dimensional data clustering tasks, feature weighting effectively captures the importance of different features within clusters. However, most mainstream clustering algorithms either lack feature selection capabilities or heavily rely on parameter tuning for feature selection, which reduces efficiency. To address these issues, this paper proposes a novel clustering algorithm based on regularization and entropy (REW-GBC). By constructing a granular-ball distribution model and employing an innovative splitting strategy, the method significantly reduces the global computational burden. Locally, we apply feature weighting within each granular-ball using regularization, assigning lower weights to less important features. Entropy is calculated to measure the certainty of the granular-ball weight determination, and the most reasonable weight for the granular-ball is selected for clustering. REW-GBC was tested on multiple real-world datasets, and experimental analysis shows that the method is highly competitive, outperforming other clustering methods in terms of effectiveness.