Guideline of Collinearity - Avoidable Regression Models on Time-series Analysis

Regression models are one of the most important machine learning models for time-series data prediction. However, multicollinearity is also an obstacle of accuracy. The most straightforward approach (called Type 1) to address the collinearity problem is to eliminate some related variables before modeling. The other approach uses new algorithms that can handle the situation well. Type II algorithms, including Linear Regression, Ridge Regression, Lasso Regression, Elastic Net Regression, Linear Regression with Polynomic features, Decision Tree Regression, and Random Forest Regression, are applied to construct a prediction model directly. The comparison between type 1 and type 2 is shown in this paper. This research aims to compare the efficiency of various regression model algorithms on datasets that consist of collinearity. The main contribution of this paper is simplicity that retains high accuracy. Elastic Net Regression offers the best choice with R square on the test data of 92%.