Improving the Accuracy of the Model Selection by Applying Optimal Tuning Parameters in the Elastic Net Penalized Quantile Regression Model and Empirical Mode Decomposition with Applications

Abstract

Selecting optimal tuning parameters can enhance the accuracy of machine learning techniques, particularly when data exhibits heterogeneity and multicollinearity. Thus, this paper introduces a novel approach by combining elastic net penalized quantile regression (QRELN) with empirical mode decomposition (EMD). The EMD algorithm is used to decompose the non-stationary and nonlinear original time series predictor into a finite set of several intrinsic mode function components and one residual component. While elastic-net quantile regression (QRELN) offers more accurate estimations by addressing multicollinearity, heavy-tailed distributions, heterogeneity, and selection of the most important variables. The results of the numerical experiments and real data confirmed the superiority of the EMD. QRELN method with selecting the optimal tuning parameters. The proposed ELNET.QR αopt method also effectively identifies predictor variables that have the most significance on the response variable.