Convergence of an EM-type algorithm for spatial clustering
Pattern Recognition Letters
An improved Bayesian structural EM algorithm for learning Bayesian networks for clustering
Pattern Recognition Letters
Opening the black box: interactive hierarchical clustering for multivariate spatial patterns
Proceedings of the 10th ACM international symposium on Advances in geographic information systems
A probabilistic framework for semi-supervised clustering
Proceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining
Genetic-Based EM Algorithm for Learning Gaussian Mixture Models
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
To consider spatial information in spatial clustering, the Neighborhood Expectation-Maximization (NEM) algorithm incorporates a spatial penalty term in the objective function. Such an addition leads to multiple iterations in the E-step. Besides, the clustering result depends mainly on the choice of the spatial coefficient, which is used to weigh the penalty term but is hard to determine a priori. Furthermore, it may not be appropriate to assign a fixed coefficient to every site, regardless of whether it is in the class interior or on the class border. In estimating class posterior probabilities, sites in the class interior should receive stronger influence from their neighbors than those on the border. To that end, this paper presents a variant of NEM using varying coefficients, which are determined by the correlation of explanatory attributes inside the neighborhood. Our experimental results on real data sets show that it only needs one iteration in the E-step and consequently converges faster than NEM. The final clustering quality is also better than NEM.