An Improved Interval Type-2 Rough Fuzzy K-means Based on Local Imbalanced Metric
Yudi Zhang, Tengfei Zhang, Fumin Ma, Xiaoyu Zhang · 2022
Because of accurately describing the uncertainty of cluster boundaries with different shapes, the interval type-2 rough fuzzy k-means clustering (IT2RFKM) has been widely used in unsupervised learning of preliminary data in recent years. Nonetheless, faced with imbalanced clusters, traditional fuzzy metric for overlapping data probably causes the cluster center of smaller cluster shifts to the boundary, which greatly interferes with clustering accuracy. In light of this, some adverse effects due to imbalanced cluster size are elucidated; A local imbalance metric called iteration center coefficient is skillfully proposed, by which the inequality of the boundary is effectively characterized, and the weighting coefficients in iterative formula of cluster center can also be adjusted adaptively. On this basis, an improved interval type-2 rough fuzzy k-means based on local imbalanced metric is further refined. The validity of proposed algorithm is demonstrated by the results of the synthetic dataset and some UCI datasets.