A1401
Title: An adaptive weighted component test for phase I monitoring in high-dimensional means
Authors: Qianwei Lan - University of Macau (China) [presenting]
Lianjie Shu - University of Macau (China)
Xinyi Ren - University of Macau (China)
Abstract: In high-dimensional Phase I process monitoring, variables frequently exhibit distinct heterogeneity in both sparsity and shift magnitude, often compounded by unequal component variances. Existing methods designed under rigid assumptions, such as extreme sparsity or pure density, suffer severe loss of power when the actual shift pattern deviates from these patterns. To bridge this gap, a nonparametric adaptive weighted component test (AWCT) control chart is proposed. The approach first normalizes component-wise statistics to mitigate variance heterogeneity. Recognizing that sparsity is essentially a special case of magnitude heterogeneity where unaffected components have shift sizes of zero, the AWCT then employs data-driven adaptive weights to simultaneously accommodate unknown sparsity and varying shift magnitudes. Extensive simulation studies demonstrate that the AWCT consistently achieves higher detection power than competing methods across diverse shift patterns. A real-world example further validates the practical value of adaptive weighting in high-dimensional monitoring.
