A Weighted Fuzzy Clustering Method Based on Granular-Ball Computing
Qiao Deng, Jiang Xie, Hongxia Hu, Minggao Dai · 2024
Existing weighted fuzzy clustering methods typically use feature weighting techniques to eliminate irrelevant features. However, their emphasis is typically on the initial points, fuzzy parameters, and cluster number of the original Fuzzy C-Means (FCM). Unfortunately, these approaches do not ensure a solution to the original FCM problem. For certain specialized applications, such as clustering “non-spherical” data, conventional weighted clustering algorithms may not demonstrate satisfactory performance. This paper proposes a weighted fuzzy clustering method based on granular-ball computing. Firstly, the relevant fuzzy granular-ball set is generated through weighted fuzzy iteration, and then the generated fuzzy granular-balls are connected by the adjacent connection rule to form the final clusters, which not only improves the clustering accuracy and other fuzzy clustering indicators, but also adapts to many special data scenarios including “non-spherical”.