Penalized multimodal mixture logit model (Q626262): Difference between revisions

A classification problem is considered where the observed classes are mixtures of some subclasses. The multinomial logit model is used to model the dependence between subclasses labels and predictors. A version of the EM algorithm is proposed for fitting the resulting mixture model. This algorithm maximizes a penalized log-likelihood, helping to avoid the overfitting problem. Results of simulations and applications to medical data are presented.