Clustering Analysis Using Data Range Aware Seeding and Agglomerative Expectation Maximization

Expectation maximization (EM) is a local maximization method of the mixture model. When applied to clustering analysis, it generates good results only with reasonably good initialization, which can be produced by hierarchical agglomeration. However, hierarchical agglomeration has poor scalability due to its computational complexity. This paper presents a novel method, called ISOEM, to overcome this limitation. It uses a data range aware seeding algorithm to create an initial classification to initialize an iterative self-organizing process. The process alternates between EM and agglomeration coupled with classification EM. Evaluation using two imagery datasets showed the method had very good performance. The paper also presents the results of using a skewness measure and a separation-cohesion index as indicators for determining the number of clusters in the data.