Enhancing Model Selection Based On Penalized Regression Methods And Empirical Mode Decomposition
In this study, the penalized regularization methods, namely, the smoothly clipped absolute deviation (SCAD), adaptive least absolute shrinkage and selection operator (adLASSO) regression, minimax concave penalty (MCP) and elastic net (ELNET) regression, are adopted. Those methods are combined with t...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/51620/1/ABDULLAH%20SULEIMAN%20SALEH%20AL%20JAWARNEH%20-%20TESIS.pdf http://eprints.usm.my/51620/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In this study, the penalized regularization methods, namely, the smoothly clipped absolute deviation (SCAD), adaptive least absolute shrinkage and selection operator (adLASSO) regression, minimax concave penalty (MCP) and elastic net (ELNET) regression, are adopted. Those methods are combined with the first part of the Hilbert–Huang transformation, namely, the empirical mode decomposition (EMD) algorithm. The EMD algorithm is employed to decompose the nonstationary and nonlinear time series dataset into a finite set of orthogonal decomposition components, which includes a set of intrinsic mode function and residual components. These components have been used in several studies as new predictor variables to predict the behaviour of the response variable. The penalized regularization methods are statistical techniques used to regularize and select the necessary predictor variables that have substantial effects on the response variable. These methods are also utilized to produce a consistent model in terms of variable selection and asymptotically normal estimates and address the multicollinearity problem when it exists between the predictor variables. This study aims to apply the proposed SCAD-EMD, adLASSO-EMD, MCP-EMD and ELNET-EMD methods to determine the effect of the decomposition components of the original univariate/multivariate time series predictor variable(s) on the response variable. Moreover, this study tackles the multicollinearity between the decomposition components to enhance the prediction accuracy for creating a fitting model. The proposed techniques are compared with four traditional regression methods employed in the previous study. |
|---|