Contemporary Regression Methods: Dealing with Non-linearity and Multicollinearity in Linear Regression

Abstract

Regression is a type predictive modelling method which forecasts, evaluates and predictive analysis, and this involves modelling future risk and opportunities. It also shows a significant relationship, and strength of impacts makes it an ideal tool in many life areas, such as the Finance and the Economic. While this article focuses to raise the understanding the use of linear regression to deal with non-linearity and multicollinearity for getting the best fit model. Non-linearity and multicollinearity attitude a serious challenge to the estimation of fit, several ways have been established to solve these challenges, such as polynomial regression and the regularization technique. Therefore, the use of polynomial regression poses a solution to fix non-linearity. Similarly, the regularization technique, which are ridge regression, Lasso regression and elastic net regression, has been established to reduce multicollinearity. The other point of this study is selecting the right model between linear regression, polynomial regression, ridge regression, Lasso regression and elastic net regression to achieve the best fit. It will also be addressing the order to have the right model between these regressions. This article suggests that the highest value of R-Square (R2), and the lowest value of the root mean square error (RMSE), can be a participant in selecting the best-fit modelling of the data.