The Impact of Multicollinearity on Big Data Multivariate Analysis Modeling

Ntotsis, Kimon; Karagrigoriou, Alex

doi:10.1002/9781119821588.ch10

Cited by 4 publications

(1 citation statement)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Linear regression analysis is used to estimate the linear relationship between the response variable and one or more independent variables for prediction purposes (e.g., [15,47]). In multiple linear regression, explanatory variables may be linearly related to each other which causes the problem of multicollinearity (e.g., [5,48,68,79]). The predictive power of the model and model fitting procedure is not much affected by the presence of multicollinearity (e.g., [55,73]).…”

Section: Introductionmentioning

confidence: 99%

Robust correlation scaled principal component regression

TAHİR>

Ilyas

2023

Hacettepe Journal of Mathematics and Statistics

View full text Add to dashboard Cite

In multiple regression, different techniques are available to deal with the situation where the predictors are large in number, and multicollinearity exists among them. Some of these approaches rely on correlation and others depend on principal components. To cope with the influential observations (outliers, leverage, or both) in the data matrix for regression purposes, two techniques are proposed in this paper. These are Robust Correlation Based Regression (RCBR) and Robust Correlation Scaled Principal Component Regression (RCSPCR). These proposed methods are compared with the existing methods, i.e., traditional Principal Component Regression (PCR), Correlation Scaled Principal Component Regression (CSPCR), and Correlation Based Regression (CBR). Also, Macro (Missingness and cellwise and row-wise outliers) RCSPCR is proposed to cope with the problem of multicollinearity, the high dimensionality of the dataset, outliers, and missing observations simultaneously. The proposed techniques are assessed by considering several simulated scenarios with appropriate levels of contamination. The results indicate that the suggested techniques seem to be more reliable for analyzing the data with missingness and outlyingness. Additionally, real-life data applications are also used to illustrate the performance of the proposed methods.

show abstract