Controlling the False Discovery Rate via Model-Agnostic Importance Measures
This study introduces Model-Agnostic Selection with agent-X importance (MAS), a new method for selecting covariates while controlling the false discovery rate (FDR). MAS bypasses distributional assumptions using a novel importance measure, a resplit-per-covariate strategy, and data aggregation. Results show MAS is competitive with existing methods in performance and computational efficiency.