Analyzing of diabetes data Using SIR-based methods
Ali Alkenani
Mohamed Abdulkadhim
Al-Qadisiyah University – College of Administration and Economics
Corresponding Author: Mohamed Abdulkadhim
Abstract : The SDR received great attention in high-dimensional regressions. Assume Y is a response variable and
X=(x_1,…,x_p )^T is a predictor of p-dimensions. Without assuming any parametric model, the main idea of SDR is
to replace X with a Low-dimensional orthogonal P_s X to S while retaining information about the Y | X distribution.
The aim of SDR procedure is to find the central subspace S_(Y|X), and that S_(Y|X)is the intersection of all
subspaces such as Y╨X|P_s X. Where ╨ denotes independence. Therefore, P_β X excerpts all the information from
X about Y, where β is the base to S_(Y|X). (Cook, 1998).
There are several proposed methods for finding S_(Y|X), and one of the well-known methods is SIR (Li, 1991),
SIR is applied in several fields including economics, and bioinformatics. SIR faces difficulties in interpreting the
resulting estimates about SIR due to its production of linear combinations from all of the original predictors. To
improve the interpretation of SIR analysis, it is necessary to decrease the number of non-zero coefficients which are
also insignificant in the SIR directions.
The objective of our study is to reduce the number of nonzero coefficients in SIR directions for obtaining better
interpretability. Through combining some of the regularization methods with the SIR method to produce sparse and
accurate estimations.
in this paper will we employ methods that merge SIR work with the Lasso method. SSIR(Ni et al, 2005), RSIR (Li
and Yin, 2008), SIR-LASSO Lin et al.( (2018) methods in analyses sample data for diabetes
