Improved Constrained k-Means Algorithm for Clustering with Domain Knowledge

Witnessing the tremendous development of machine learning technology, emerging applications impose challenges using domain knowledge to improve accuracy clustering provided that suffers a compromising rate despite its advantage fast procession. In this paper, we model (i.e., background or side information), respecting some as must-link and cannot-link sets, for sake collaborating with k-means better accuracy. We first propose an algorithm constrained k-means, considering only must-links. The key idea is consider set data points by must-links single point weight equal sum points. Then, cannot-link, employ minimum-weight matching assign existing clusters. At last, carried out numerical simulation evaluate proposed algorithms against UCI datasets, demonstrating our method outperforms previous well traditional regarding although slightly compromised practical runtime.