Mixture Models

7Robust mixture regression models

DOI: 10.1201/9781003038511-7

As explained in Section 3, the unknown mixture regression parameters are usually estimated by maximum likelihood estimators (MLE) using the EM algorithm based on the normality assumption of component error density. Similar to the ordinary least-squares estimates for the linear regression, the normal mixture linear regression can be very sensitive to the presence of gross outliers or heavy-tailed error distributions, failing to accommodate for the outlying effects greatly jeopardizing both model estimation and inference. In fact, even a single atypical value may have a large effect on the parameter estimators. To overcome this problem, many robust methods for mixture regression ...