Fuzzy Dissimilarity Measure Based K-Means Clustering

The k-means algorithm is a classic unsupervised learning approach for the clustering problem, which has a good effect on processing spherical data. Since its initial centroids are randomly assigned, the clustering effect is usually not stable. Moreover, the traditional k-means and related methods often only consider the similarity of the instances to each cluster’s centroid, therefore such methods are easily affected by noise points and hard to solve linearly indivisible data. In this paper, a novel clustering algorithm is proposed to improve the processing power of complex data distribution using the fuzzy dissimilarity of instances in different clusters. Experimental results confirm that the proposed approach works effectively and generally outperforms the popular representative algorithms for both artificial and benchmark datasets.