A Daubechies wavelet transformation to optimize modeling calibration of active compound on drug plants

In modeling calibration, a dependency among predictor variables becomes major problem resulting in a not unique given model parameter estimation. This problem may degrade the performance value of the calibration model. A concentration of drug's active compound in 20 locations involving the proceeds percent transmittance observed at 1866 wavelengths as a predictor variable. The magnitude of the dimensions of the predictor variables do not guarantee the independency between variables. PCA or PLS become a mainstay method from several researches to overcome dependency by reducing the dimensions of the predictor variables. The information reduction process by wavelet becomes a major concern. A Daubechies wavelet is able to caover a polynomial trend, while the Haar wavelet is discontinuous function. In this paper, we investigate the Daubechiese and Haar wavelet perfomance to handle multi-dependency in case of overdimension on calibration model. Moreover, the performance of Haar and Daubechies may be assessed through the RMSEP regression model. The dimension reduction process through Daubechies wavelet provides better prediction accuracy of calibration models than PCA or PLS.