A new semi-supervised fuzzy K-means clustering method with dynamic adjustment and label discrimination

Conventional unsupervised Fuzzy K-means methods (FKM) usually analyze the structure of data solely without considering the influence of label information carried by data, which limits the performance and stability of clustering. How to leverage annotated label information to improve the performance of unsupervised FKM methods is still a challenging research problem. To that end, this paper proposes a new Semi-Supervised Fuzzy K-means method (SSFKM) consisting of dynamic adjustment and label discrimination. Specifically, dynamic adjustment aligns label information and clustering results to distinguish the learning difficulties of labeled data and enable the method to focus on simple but reliable label information. Moreover, a new distance measure is designed to re-evaluate the membership of labeled data with cluster centers, forcing labeled data to be classified into correct cluster for enhancing label discrimination. Comprehensive experiments demonstrate that the SSFKM method achieves the best performance compared with existing state-of-the-art semi-supervised clustering methods. In addition, the results demonstrate that the SSFKM method reduces the impact of data noise effectively during clustering.