Pseudo-univariate calibration based on near infrared spectroscopy and independent component analysis

Independent component analysis (ICA) searches for the decomposition of signals of a mixture into most statistically independent components. It is based on the construction of independent components (ICs) which are linear combinations of the original variables. The ICs are assumed to correspond to the signals of the pure source signals present in the analyzed mixtures and its respective scores. The hypothesis used to enable the extraction of the pure source signals is that these vectors are statistically independent, as opposed to principal component analysis, which is based on calculating orthogonal vectors that maximize the amount of variance extracted from the data. In this work is shown that ICA can be used to pseudo-univariate calibration from near infrared spectroscopy (NIR) data in two ways to determine carbendazim concentration in orange juice. To this, the orange juice from a safe source (without carbendazim) was spiked with carbendazim in the concentration range 0.5 to 2.0 (mg/mL). The NIR spectra (900 – 1700nm) were acquired in a JDSU MicroNIR® equipment. The ICA was performed using for Matlab® with JADE (Joint Approximate Diagonalization of Eigenmatrices) algorithm after first derivative to baseline correction. In the first case, the scores from ICA were plotted against carbendazim concentrations in a pseudo-univariate calibration model with correlation coefficient 0.9791. By using this model, the determination of carbendazim concentrations, in two independent samples, showed absolute error of 0.1 and 0.22 mg/mL. On the second case, the scores from ICA were used to perform a first order standard addition method. In this application, six samples were used (the first sample was considered an original sample plus five standard additions). In the quantification procedure, the analytical curve of standard addition was built plotting the scores from ICA against the carbendazim concentration. The result was a correlation coefficient of 0.9827 and carbendazim concentration 0.6 mg/mL, which implies in a good agreement with the expected value (0.5 mg/mL). Then, it can be concluded that the combination of NIR measurements and ICA makes possible the direct quantification of carbendazim in matrices with unknown interferents. Therefore, for both examples, it is possible to conclude that ICA can provide the second order advantage with first order data.