A Classification EM algorithm for clustering and two stochastic versions
Computational Statistics & Data Analysis - Special issue on optimization techniques in statistics
Convergence of an EM-type algorithm for spatial clustering
Pattern Recognition Letters
A view of the EM algorithm that justifies incremental, sparse, and other variants
Learning in graphical models
Clustering Ensembles: Models of Consensus and Weak Partitions
IEEE Transactions on Pattern Analysis and Machine Intelligence
Hi-index | 0.00 |
Like other iterative refinement clustering algorithms, the Neighborhood Expectation-Maximization (NEM) algorithm is sensitive to the initial state of cluster separation. Therefore, the study of the initialization methods is of great importance for the success of finding a better sub-optimal solution in practice. However, existing initialization methods for mixture model based clustering using EM-style algorithms do not account for the unique properties of spatial data, such as spatial autocorrelation. To that end, this paper incorporates spatial information into the initialization and compares three representative initialization methods. Our experimental results on both synthetic and real-world datasets show that the NEM algorithm usually leads to a better clustering solution if it starts with initial states returned by the spatial augmented initialization method based on K-Means.