Performances Comparison of Information Criteria for Outlier Detection in Multiple Regression Models Having Multicollinearity Problems using Genetic Algorithms

Abstract

Multiple linear regression models are widely used in applied statistical techniques and they are most useful devices for extracting and understanding the essential features of datasets. However, in multiple linear regression models, problems arise when multicollinearity or a serious outlier observation present in the data. Multicollinearity is a linear dependency between two or more explanatory variables in the regression models which can seriously affect the least squares estimated regression surface. The other important problem is outlier; they can strongly influence the estimated model, especially when using least squares method. Nevertheless, outlier data are often the special points of interests in many practical situations. The purpose of this study is to performance comparison of Akaike Information Criterion (AIC'), Bayesian Information Criterion (BIC’) and Information Complexity Criterion (ICOMP'(IFIM)) for detecting outliers using Genetic Algorithms when multiple regression model having multicollinearity problems. Keywords: Akaike Information Criterion; Bayesian Information Criterion; Information Complexity Criterion; Genetic Algorithms; multicollinearity; outlier detection. 2010 Mathematics Subject Classification: 62J05