Feature selection based on statistical estimation of mutual information

An algorithm to identify significant factors is proposed in the mixed model framework. It employs statistical estimation of mutual information. Consistency of this procedure is established. Numerical experiments demonstrating its accuracy supplement theoretical results.