Implementasi Algoritma EM pada Metode Kemungkinan Maksimum untuk Pemodelan Regresi Linear Gerombol

Abstract

Clusterwise regression modelling consider the several hidden clusters from a data set which have different regression functions. This method is used simultaneously to determine the number of clusters, to separate membership into specified cluster K, and to estimate each regression function. Maximum likelihood methodology implemented by Expectation-Maximization (EM) algorithm is used for parameter estimation. EM algorithm consists of two steps. The first step is expectation (E-step), to count log-likelihood function, and the second step is maximization (M-step), to determine the new parameter value which maximizes log-likelihood function. The best regression coefficients estimation and the number of optimal clusters are obtained when log-likelihood value is maximum and Akaike’s Information Criterion (AIC) value is minimum. Some simulation data sets in this research are provided with some criteria that combined with fractional factorial design.