Latent class regression model in IRLS approach

We consider latent class regressions for the simultaneous construction of several regression models by the data clusters. Maximum likelihood objective of observations belonging to at least one data segment is developed. Solution is reduced to the iteratively reweighted least squares (IRLS) procedure that defines coefficients of all models and the characteristics of fitting. Together with the regression models, this approach yields probabilities of each observation belonging to each of the classes. This technique can also be used for finding parameters of mixed distributions. The suggested approach enriches results of the regression modeling and clustering in practical applications.