An improved preconditioned unsupervised K-means clustering algorithm

The Unsupervised K-means clustering (UKM) algorithm has attracted the attention of many researchers because it can automatically identify the number of clusters without requiring any parameter selection. However, it may produce poor clustering results on datasets with Gaussian mixtures. In this paper, we consider the preconditioned UKM algorithm, where the truncated UKM algorithm is first used as a preconditioning strategy. To further enhance the algorithm's performance, we introduce a circular modification strategy. In particular, we determine whether to use the above strategies based on the Bayesian Information Criterion (BIC). The experimental results reveal that the proposed algorithms have a higher clustering accuracy than the UKM algorithm when applied to Gaussian mixture datasets.