Supervised learning of multivariate skew normal mixture models with missing information (Q2430235)

Properties of the multivariate skew normal mixture (MSNMIX) model in a missing information framework are studied. The proposed model is flexible in dealing with heterogeneous data that involve strong skewness and is robust to the presence of missing observations. An analytically tractable EM algorithm coupled with some useful model-based tools to handle data with general missing patterns in the class of MSNMIX models is offered. This algorithm is used to compute the ML estimates from incomplete data. Statistical principles regarding classification and prediction of incomplete features are investigated. The proposed methodologies are applied to a real data set with varying proportions of synthetic missing values.