A Feature-Clustering Algorithm Based on K-Medoids

Clustering is an important problem in machine learning. Existing clustering methods mainly focus on sample clustering, while research on feature clustering remains limited. Unlike sample clustering, features exhibit substantial heterogeneity in their types and distributions, rendering traditional clustering algorithms based on homogeneity assumptions ineffective. To address this issue, this work proposes a novel feature-clustering algorithm based on k-medoids (FCKM). It adapts three distance metrics according to different types of features and employs an improved k-medoids algorithm on feature clustering. Besides, this work proposes a novel metric to evaluate the performance of feature clustering. Experimental results on real-world datasets show that our proposed algorithm has better performance than its competitor, showcasing strong practical applicability.