Multi-resolution analysis of near infrared spectroscopic data for calibration and prediction of active substances in phosphate buffer solution

Abstract

Discrete wavelet transform is performed on near infrared spectroscopic data as a preprocessing step in calibration and prediction of active substance concentrations in phosphate buffer solutions. Wavelet coefficients from different scales are used as predictor variables to build regression model via partial least squares. The prediction results are compared with calibration model developed from raw spectroscopic data (without wavelet transform). The wavelet scale, which gives the best-improved prediction accuracy, is selected to form prediction model.