A proposed hybrid method for Multivariate Linear Regression Model and Multivariate Wavelets (Simulation study)

Resumen

Abstract: In this paper, a hybrid method was proposed for multivariate linear regression model and multivariate wavelets. Including the wavelet transform for the dependent variables through the multivariate Daubechies and Fejer-Korovkin wavelets and using the minimax and universal methods with the soft threshold rule to data de-noise when estimating model parameters. Then the comparison between the proposed hybrid and classical method (Ordinary Least Squares), combining simulated and actual data along with a MATLAB program written specifically for this purpose. The best possible multivariate linear regression model for the data may be obtained based on the mean squared error. The research showed that the proposed hybrid method yields more accurate parameter estimates than the traditional approach.

Keywords: Multivariate Wavelet, multivariate linear regression model, De-noise, and threshold.