An self-adaptive cluster centers learning algorithm based on expectation maximization algorithm

Abstract

It is called unsupervised learning that does not rely on any labeled value, and finds the relationship between samples by mining the intrinsic characteristics of samples. Clustering algorithm is a kind of unsupervised learning algorithm. Although many clustering algorithms have been studied in modern science and applied in many fields, it is their common problem that the quantity of clusters has to be specified. Based on EM algorithm, this paper proposes a cluster centers learning algorithm (CCL) which can self-adaptively calculate the quantity and parameters of clusters according to the characteristics of samples themselves. The algorithm tentatively fills the shortage of existing clustering algorithms. The paper proposes the elementary merger and splitting criteria. The criteria can determine whether a point is the cluster center according to the characteristics of samples. Based on the elementary criteria, the algorithm proposed by the paper can adapt to calculate the correct quantity of clusters and gives the corresponding clustering parameters. Monte Carlo simulation is used to evaluate the effectiveness of the proposed algorithm. The experimental results show that the algorithm proposed by the paper can start from an arbitrary given cluster center and calculates the cluster centers close to the actual cluster centers of the samples themselves, so as to complete the self-adaptive unsupervised clustering.