Investigating the factors which affect the performance of the EM algorithm in Latent class models

Abstract

Latent class models have been used extensively in market segmentation to divide a total market into market groups of consumers who have relatively similar product needs and preferences. The advantage of these models over traditional clustering techniques lies in simultaneous estimation and segmentation, which is carried out using the EM algorithm. The identification of consumer segments allows target-marketing strategies to be developed. The data comprises the rating responses of 262 respondents to 24 laptop profiles described by four item attributes including the brand, price, random access memory (RAM) and the screen size. Using the facilities of R Studio, two latent class models were fitted by varying the number of clusters from 2 to 3. The parameter estimates obtained from these two latent class models were used to simulate a number of data sets for each cluster solution to be able to conduct a Monte-Carlo study, which investigates factors that have an effect on segment membership and parameter recovery and affect computational effort.