A Recursive Factor Elimination Method for the Analysis of Supersaturated Designs With Binary Response

Supersaturated designs are characterized by a number of factors that exceeds the available runs, resulting in insufficient degrees of freedom to estimate all main effects. Consequently, analyzing these designs presents a substantial challenge. Such designs are frequently employed in screening experiments, where the primary objective is to identify the dominant active factors while minimizing cost. Although numerous methods have been developed to analyze supersaturated designs with a normally distributed dependent variable, relatively few have addressed the case of a non‐normal response. Recent advancements in statistical learning algorithms have spurred the development of novel approaches. To this end, in this work, we introduce a new methodology for identifying key factors in supersaturated designs with a Bernoulli‐distributed response. The proposed method consists of two steps: it integrates support vector classification in a modified recursive structure with penalized logistic regression via the LASSO. This approach, named recursive support vector classification with LASSO (RSVC‐L), effectively detects active effects in both two‐level and mixed‐level supersaturated designs. Findings from an extensive simulation study indicate that RSVC‐L exhibits high statistical power and low Type I error rates, establishing it as both a robust and effective technique.