INVESTIGADORES
GOICOECHEA Hector Casimiro
artículos
Título:
Sustained prediction ability of net analyte preprocessing methods using reduced calibration sets. Theoretical and experimental study involving the spectrophotometric determination of multicomponent mixtures
Autor/es:
GOICOECHEA, HÉCTOR C; AC OLIVIERI,
Revista:
ANALYST
Editorial:
RSC
Referencias:
Año: 2001 vol. 126 p. 1105 - 1113
ISSN:
0003-2654
Resumen:
A newly developed multivariate method involving net analyte preprocessing (NAP) was tested using central
composite calibration designs of progressively decreasing size regarding the multivariate simultaneous
spectrophotometric determination of three active components (phenylephrine, diphenhydramine and naphazoline)
and one excipient (methylparaben) in nasal solutions. Its performance was evaluated and compared with that of
partial least-squares (PLS-1). Minimisation of the calibration predicted error sum of squares (PRESS) as a
function of a moving spectral window helped to select appropriate working spectral ranges for both methods. The
comparison of NAP and PLS results was carried out using two tests: (1) the elliptical joint confidence region for
the slope and intercept of a predicted versus actual concentrations plot for a large validation set of samples and
(2) the D-optimality criterion concerning the information content of the calibration data matrix. Extensive
simulations and experimental validation showed that, unlike PLS, the NAP method is able to furnish highly
satisfactory results when the calibration set is reduced from a full four-component central composite to a fractional
central composite, as expected from the modelling requirements of net analyte based methods.versus actual concentrations plot for a large validation set of samples and
(2) the D-optimality criterion concerning the information content of the calibration data matrix. Extensive
simulations and experimental validation showed that, unlike PLS, the NAP method is able to furnish highly
satisfactory results when the calibration set is reduced from a full four-component central composite to a fractional
central composite, as expected from the modelling requirements of net analyte based methods.