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
Fast spatial clustering with different metrics and in the presence of obstacles
Proceedings of the 9th ACM international symposium on Advances in geographic information systems
Opening the black box: interactive hierarchical clustering for multivariate spatial patterns
Proceedings of the 10th ACM international symposium on Advances in geographic information systems
CLARANS: A Method for Clustering Objects for Spatial Data Mining
IEEE Transactions on Knowledge and Data Engineering
Spatial Clustering in the Presence of Obstacles
Proceedings of the 17th International Conference on Data Engineering
Genetic-Based EM Algorithm for Learning Gaussian Mixture Models
IEEE Transactions on Pattern Analysis and Machine Intelligence
A model for non-parametric spatially varying regression effects
Computational Statistics & Data Analysis
| Hi-index | 0.00 |
The Neighborhood Expectation-Maximization (NEM) algorithm is an iterative EM-style method for clustering spatial data. Unlike the traditional EM algorithm, NEM has the spatial penalty term incorporated in the objective function. The clustering performance of NEM depends mainly on two factors: the choice of the spatial coefficient, which is used to weigh the penalty term; and the initial state of cluster separation, to which the resultant clustering is sensitive. Existing NEM algorithms usually assign an equal spatial coefficient to every site, regardless of whether this site is in the class interior or on the class border. However, when estimating posterior probabilities, sites in the class interior should receive stronger influence from its neighbors than those on the border. In addition, initialization methods deployed for EM-based clustering algorithms generally do not account for the unique properties of spatial data, such as spatial autocorrelation. As a result, they often fail to provide a proper initialization for NEM to find a good solution in practice. To that end, this paper presents a variant of NEM, called ANEMI, which exploits an adaptive spatial coefficient determined by the correlation of explanatory attributes inside the neighborhood. Also, ANEMI runs from the initial state returned by the spatial augmented initialization method. Finally, the experimental results on both synthetic and real-world datasets validated the effectiveness of ANEMI.