Statistically equivalent feature subsets:
https://www.jstatsoft.org/article/view/v080i07

Paul Murtaugh on why AIC-based model selection is similar to just using P-values: doi.org/10.1890/13-1858.1