An Extension of the GDS-ARM Algorithm for Factor Screening in Mixed-Level Supersaturated Designs

Abstract Gauss-Dantzig Selector-Aggregation over Random Models (GDS-ARM) algorithm is a powerful method for screening main-effects and two-factor interactions in supersaturated designs. Its application, however, has been limited to two-level designs, and its critical tuning parameters are not optimized for the more complex mixed-level context. This paper provides a systematic tuning guide for extending GDS-ARM to mixed-level designs. Through extensive simulations using a diverse set of designs, we sequentially analyze the four key tuning parameters of GDS-ARM. Additionally, we identify a critical characteristic of the design matrix, the ratio of main-effect columns to runs, and provide a practical guideline for design selection based on this metric. The robustness of this complete set of recommendations is confirmed through a validation study on a large design with 140 runs.