Modified versions of the Bayesian Information Criterion for sparse Generalized Linear Models

摘要: Abstract The classical model selection criteria, such as the Bayesian Information Criterion (BIC) or Akaike information criterion (AIC), have a strong tendency to overestimate number of regressors when search is performed over large potential explanatory variables. To handle problem overestimation, several modifications BIC been proposed. These versions rely on supplementing original with some prior distributions class possible models. Three are presented and compared in context sparse Generalized Linear Models (GLMs). related choices priors discussed conditions for asymptotic equivalence these criteria provided. performance modified illustrated an extensive simulation study real data analysis. Also, simplified BIC, based least squares regression, investigated.