Application of a Partial Least Squares Regression for the Determination of Nanomolar Concentrations of Scandium in the Presence of Nickel by Adsorptive Stripping Voltammetry

This paper presents the methodology of a very sensitive determination of scandium in excess of nickel by adsorptive stripping voltammetry on a mercury film electrode and PLS regression. A calibration set consisting of binary mixtures containing 5, 15, 25, 35 or 45x10(-9)M Sc(III) and simultaneously 0.5-50x10(-7)M of Ni(II) was used to develop the chemometric PLS calibrations. An external set containing synthetic mixtures of 10, 20, 30, 40x10(-9)M Sc(III) and the same Ni(II) concentration as mentioned above was used to validate the model and evaluate predictive ability. The application of data pretreatment techniques involving baseline correction, smoothing, range-scaling, mean-centering and their influence on the PLS model complexity, were also investigated. In the effect, the model for Sc(III), including 6 latent variables, was constructed. The model fulfills validation criteria and is characterized by a good prediction ability (majority of the prediction errors are lower than 10%). This work shows significant progress in the development of a very sensitive analytical technique for the determination of scandium in the presence of different concentrations of nickel by application of multivariate calibration tools.