Abstract:

As system complexity increases， identifying key factors that significantly affect quality characteristics with minimal experimental cost becomes a critical step in quality improvement activities. However， challenges such as limited samples， economic constraints of experimentation， and the presence of unbalanced data necessitate the development of new factor screening methods. A large-scale factor screening approach based on unbalanced sequential design is introduced to address these issues.First， a first-order model that simultaneously considers both location and dispersion effects of factors is constructed. Integrating the fundamental assumptions and framework of sequential bifurcation （SB）， the SB-UB method is proposed to handle two types of unbalanced data. For the first type of unbalanced data， an improved Bradley-Blackwood test is introduced， while for the second type， a dual test combining F-test and Student’s t-test is proposed. Both methods aim to examine the significance of the location and dispersion effects of the factors. Monte Carlo simulation experiments demonstrate the effectiveness and robustness of the proposed SB-UB method for large-scale factor screening. By incorporating both location and dispersion effects， this approach enhances the accuracy of identifying critical quality factors while maintaining a low experimental cost.To validate the method， several simulation experiments， including large-scale simulation systems such as supply chain models， are presented. These systems often involve dozens or even hundreds of factors， far exceeding the capabilities of traditional factor screening methods， which are typically designed for problems involving fewer than 20 factors. The proposed method allows for effective screening even when unbalanced data， due to the use of multiple devices with differing computational capacities， impacts the quality of experimental results. The primary results of this research contribute to the field of quality improvement by providing a robust method to identify key factors under challenging conditions of unbalanced and large-scale data. The integration of advanced statistical testing techniques into the SB framework significantly improves the ability to detect both location and dispersion effects. Furthermore， a gap in the literature is filled by addressing the unbalanced data problem in factor screening， which has been largely overlooked in the field of simulation-based system analysis.In conclusion， the proposed SB-UB method not only advances the sequential bifurcation approach but also provides practical solutions for large-scale factor screening under unbalanced conditions， aiding engineers and researchers in making informed decisions about critical quality factors. The research findings are expected to have broad applications in areas such as supply chain management， service science， and various engineering fields.

Key words: unbalanced design, large-scale factors, factor screening, sequential bifurcation method, dispersion effects