Robust variable selection for finite mixture regression models (Q1753969)

The problem of simultaneous estimation and variable selection in finite mixture models is considered. Specifically, a procedure based on minimum-distance techniques (namely, squared Hellinger distance) is proposed. The proposed estimator has the variable selection consistency and oracle property. Moreover, the global robustness properties of the proposed estimation are established by finding its finite-sample breakdown point. A Monte Carlo study is also performed together with a real data application.