EVALUATION OF THE PERFORMANCE OF WAVELETS FOR MULTIVARIATE REGRESSION MODELS USING INFRARED SPECTROSCOPY DATA

Authors

• Marcia Werlang
• Marlize Rizzi
• Eloisa Marciana Kolberg Theisen
• Germano Fenner
• Marco Flôres Ferrão
• Annelise Engel Gerbase
• Rubén Panta Pazos

DOI:

https://doi.org/10.17058/tecnolog.v12i1.126

Keywords:

PLS, otimização, wavelets, redução da dimensionalidade, infravermelho, polióis de soja

Abstract

Discrete wavelet transform (DWT) Daubecheis was used to compress the dimension of spectral infrared data for determination to the hydroxyl value (OHV) of soybean polyols samples. Spectral data were recorded between 650 and 4000 cm-1 with a 4 cm-1 resolution by Fourier transform infrared spectroscopy (FTIR) coupled with attenuated total reflection (ATR) accessory. Through the models of regression using partial least squares (PLS) and interval partial least squares (iPLS) methods, the performance of each was compared with the original and/or between them. The spectra data set compressed the 1/4 of its original dimension they had presented the best one resulted with a lesser RMSEP that the model with the not compress signal and a similar correlation. With this result a model of lesser dimension was gotten however with the same capacity, thus DWT, getting a robust method for the reduction of the dimension of the spectra data sets, when if to intend to construct regression multivariate models.