Student Sliced Inverse Regression

Alessandro Chiancone (1, 2, 3) , Florence Forbes (2) , Stéphane Girard (2)

1 GIPSA-SIGMAPHY - GIPSA - Signal Images Physique
2 MISTIS - Modelling and Inference of Complex and Structured Stochastic Systems
3 TU Graz - Graz University of Technology [Graz]

Sliced Inverse Regression (SIR) has been extensively used to reduce the dimension of the predictor space before performing regression. Recently it has been shown that this techniques is, not surprisingly, sensitive to noise. Different approaches has been proposed to robustify SIR, in this paper we start considering an inverse problem proposed by R.D. Cook and we show that the framework can be extended to take into account a non-Gaussian noise. Generalized Student distribution are considered and all parameters are estimated via EM algorithm. The algorithm is outlined and tested comparing the results with different approaches on simulated data. Results on a real dataset are provided showing the interest of this technique in presence of outliers.

Mots clés: dimension reduction; outliers; robust estimation

Domaines: Statistiques [math.ST]

Alessandro Chiancone, Florence Forbes, Stéphane Girard. Student Sliced Inverse Regression. 2016. ⟨hal-01294982v1⟩