Neural Networks
FREM: fast and robust EM clustering for large data sets
Proceedings of the eleventh international conference on Information and knowledge management
On convergence properties of the em algorithm for gaussian mixtures
Neural Computation
Hi-index | 0.00 |
We present a comparative study between Expectation-Maximization (EM) trained probabilistic neural network (PNN) with random initialization and with initialization from Global k-means. To make the results more comprehensive, the algorithm was tested on both homoscedastic and heteroscedastic PNNs. Normally, user have to define the number of clusters through trial and error method, which makes random initialization to be of stochastic nature. Global k-means was chosen as the initialization method because it can autonomously find the number of clusters using a selection criterion and can provide deterministic clustering results. The proposed algorithm was tested on benchmark datasets and real world data from the cooling water system in a power plant.