The adaptive LASSO regression and empirical mode decomposition algorithm for enhancing modelling accuracy.

The first part of the Hilbert–Huang transformation is named the empirical mode decomposition (EMD). Which employed to decompose the non-stationary and non-linear time series dataset into a finite set of orthogonal decomposition components. These components have been used in several studies as the new predictor variables to predict the behavior of the response variables. Adaptive LASSO (AdLASSO) regression is a technical penalized regression method used to determine the most relevant predictors on the response variable with achieving the consistency in terms of variable selection and ensuring that they are asymptotically normal. Hence, the main objective of this study is to apply the proposed EMD-AdLASSO method involving two cases of initial weights to identify the decomposed components that exhibit the strongest effects to produce a consistent model and to improve the prediction accuracy. The simulation study and real dataset used the daily exchange rate dataset of three countries against the US dollar are applied. The results showed that the proposed method in the two cases of the initial weight outperformed other existing methods by effectively identifying the decomposition components, with high prediction accuracy. This is primarily observed in the case of using the ridge regression method based on the EMD as the initial weight in the proposed EMD-AdLASSO method. [ABSTRACT FROM AUTHOR]