Groupwise sufficient dimension reduction via conditional distance clustering (Q2303753)

Dimension reduction has become a significant issue in high dimensional data analysis. The sufficient dimension reduction has been studied by \textit{K.-C. Li} [J. Am. Stat. Assoc. 86, No. 414, 316--327 (1991; Zbl 0742.62044)], and the goal of sufficient dimension reduction is to estimate \(S_{Y|X}\) and minimal dimension, where \(S_{Y|X}\) is the central subspace. In this paper, a groupwise sufficient dimension reduction framework via conditional distance clustering, when the grouping information is unknown, has been constructed. A clustering process, identify the group structure and conduct subsequent dimension reduction has been formulated. Both simulations and a real data analysis demonstrate that the clustering strategy is effective.