engleski

Cluster Detection in Noisy Environment by Using the Modified EM Algorithm

The paper studies the problem of cluster detection in noisy environment. The solution of this problem is based on the well known Expectation Maximization (EM) algorithm. By utilizing the Mahalanobis distance, and modifying the hidden variable, the rejection procedure is constructed so that it omits data from calculation of the current iteration step. Thus we construct the adaptive framework for solving the above problem. Several numerical examples are presented to illustrate the proposed algorithm.

clustering ; EM ; Mahalanobis distance ; least squares ; least absolute deviation ; Davies-Bouldin index

nije evidentirano