Inverse regression method in data structure analysis (Q1198609)

The authors present an analysis method for exploring multivariate data of high dimensionality and presenting a nonlinear structure. This method is based on the projection pursuit (PP) method of authors like Huber, Friedman and Hall, improved by the inversion regression technique of \textit{K.-C. Li} [see J. Am. Stat. Assoc. 86, No. 414, 316-342 (1991; Zbl 0742.62044)]. The proposed method, which results from the minimization of the residual sum of squares approach, claims reduction of computer time w.r.t. the \(PP\)-method and calculation simplification w.r.t. Li's method. Furthermore it is proven that under certain conditions the estimators of the parameters of the projective direction, are statistically consistent. The presented results seem interesting for all who are confronted with this problem from a theoretical point of view: the problem is well introduced and compactly handled. The practitioner may however regret not to see any examples treated nor to get actual results confirming the claims of reduced computer time and the calculation simplifications.