Nonlinear multivariate calibration methods in analytical chemistry

Calibration of microhotplate conductometric gas sensors by non-linear multivariate regression methods

“…This would theoretically enhance selectivity to individual target analytes. Similar thermal modulation has been applied to arrays of semi-selective piezoelectric sensors [36] and arrays of Taguchi gas sensors [27,37]. However, the optimal calibration of the microhotplate sensors is potentially more difficult to achieve and, consequentially, this work focuses on calibration of steady state analysis of MHP sensors.…”

“…Locally weighted regression (LWR) [23][24][25][26][27] uses a subset of the calibration data to create a linear model around each unknown sample analyzed. Alternating conditional expectations (ACE) is a global transformation regression method.…”

Sustained modelling ability of artificial neural networks in the analysis of two pharmaceuticals (dextropropoxyphene and dipyrone) present in unequal concentrations

“…PLS is a standard method for analysing multicomponent mixtures, which has been systematically applied in pharmaceutical preparations. The presence of certain types of mild non-linearities can in principle be modelled by PLS using additional spectral factors [7,8,9]. In order to apply this and other multivariate methods, electronic absorption spectra for the standard samples were recorded in the range 200-310 nm.…”

“…Although PLS assumes a linear relationship between the measured sample parameters and the intensity of its absorption bands, several authors have postulated that small deviations from linearity are acceptable as they can readily be suppressed by including additional modelling factors [7,8,9]. However, in the presence of substantial non-linearity, PLS tends to give large prediction errors and calls for more suitable models.…”

“…However, in the presence of substantial non-linearity, PLS tends to give large prediction errors and calls for more suitable models. Intrinsically non-linear calibration techniques such as non-linear partial leastsquares (NPLSs), locally weighted regression (LWR), alternating conditional expectations (ACE) [9] and artificial neural networks (ANNs) [10,11,12,13,14,15,16,17,18] are applicable in the latter cases. However, it is important to state that these methodologies should be primarily used when a data set is known or suspected to be non-linear [13].…”