关键词: K均值聚类, 粒计算, 粒距离, 无监督学习, 粒聚类

Abstract: K-means clustering belongs to unsupervised learning, which has the advantages of simple application, strong interpretability and a good clustering effect. However, it has slow convergence speed, parameters are difficult to determine, and it is easy to fall into a local solution. In order to overcome the inherent defects of K-means clustering, combined with granular computing theory, this paper proposes a new clustering model: granular K-means clustering algorithm. Samples are granulated into granules on a single feature, and the granules on multi-dimensional features forms a granular vector. Several granular distances are further defined to measure the distance between granules. Then, a granular K-means clustering method based on the granular distance is proposed, and its clustering algorithm is also designed. The granulation aims to compare the similarity in all sample spaces, which reflects the global characteristics of samples and makes the clustering converge easily with fewer iterations. Finally, experiments are carried out on several biological datasets to compare the convergence speed, K-value influence and clustering effect. The results show that the proposed granular K-means clustering method has the advantages of fast convergence speed and a good clustering effect.

Key words: K-means clustering, granular computing, granular distance, unsupervised learning, granular clustering