KINERJA REGRESI PROSES GAUSSIAN UNTUK PEMODELAN KALIBRASI PEUBAH GANDA PADA DAERAH IDENTIFIKASI SPEKTRA INFRA MERAH SENYAWA AKTIF

Abstract

Multivariate calibration models have been developed usually by using principal component regression and partial least squares regression. This research proposes the application of Gaussian process regression as an alternative method to develop a calibration model. This method is applied to the measurement of curcumin concentration based on FTIR spectra. FTIR spectra was choosen at the identification are of curcuminoid. To handle the high dimensionality of spectra data, principal component analysis was initially performed, followed by applying the Gaussian process regression. Using three principal components, 99,66% of the original data’s variability can be explained. This model was attempted for various covariance functions. The results indicate that the most relevant and suitable covariance function for curcumin concentration measurement was Matern 3 isotropic. The hyperparameter values for Matern 3 isotropic were estimated by Maximum Marginal Likelihood Method. Based on RMSEP criteria, the performance of Gaussian process regression with covariance function Matern 3 isotropic at infra red spectra indentification area is better than at wavelength number 4000-400 〖cm〗^(-1). Keywords: Gaussian Process Regression, Covariance Function, Hyperparameter, Maximum Marginal Likelihood Function.