A COMPARATIVE STUDY ON ROBUST REGRESSION METHODS

Abstract: The research should analyse data after removing the outliers which have high influences or reducing the effects of influential points. The paper introduces the robust estimation methods to reduce the influences of outliers in regression modelling. We describe LTS estimator, LMS estimator, M-estimator, S-estimator, and MM-estimator among various robust estimation methods. Then, we make an experiment for real data and investigate the performances for several methods. The result shows that the robust estimation methods with reduction of influential points perform better than ordinary least squares method in regression analysis.

Keywords: Robust regression, Influential point, Least trimmed of squares, MM-estimator