Error-Based Knockoffs Inference for Controlled Feature Selection

Recently, the scheme of model-X knockoffs was proposed as a promising solution to address controlled feature selection under high-dimensional finite-sample settings. However, procedure depends heavily on coefficient-based importance and only concerns control false discovery rate (FDR). To further improve its adaptivity flexibility, in this paper, we propose an error-based knockoff inference method by integrating features, statistics, stepdown together. The does not require specifying regression model can handle with theoretical guarantees controlling proportion (FDP), FDR, or k-familywise error (k-FWER). Empirical evaluations demonstrate competitive performance our approach both simulated real data.