GENETIC ALGORITHM-BASED PROTOCOL FOR COUPLING DIGITAL FILTERING AND PARTIAL LEAST-SQUARES REGRESSION - APPLICATION TO THE NEAR-INFRARED ANALYSIS OF GLUCOSE IN BIOLOGICAL MATRICES
Re. Shaffer et al., GENETIC ALGORITHM-BASED PROTOCOL FOR COUPLING DIGITAL FILTERING AND PARTIAL LEAST-SQUARES REGRESSION - APPLICATION TO THE NEAR-INFRARED ANALYSIS OF GLUCOSE IN BIOLOGICAL MATRICES, Analytical chemistry, 68(15), 1996, pp. 2663-2675
A multivariate calibration procedure is described that is based on the
use of a genetic algorithm (GA) to guide the coupling of bandpass dig
ital filtering and partial least-squares (PLS) regression, The measure
ment of glucose in three different biological matrices with near-infra
red spectroscopy is employed to develop this protocol, The GA is emplo
yed to optimize the position and width of the bandpass digital filter,
the spectral range for PLS regression, and the number of PLS factors
used in building the calibration model, The optimization of these vari
ables is difficult because the values of the variables employ differen
t units, resulting in a tendency for local optima to occur on the resp
onse surface of the optimization, Two issues are found to be critical
to the success of the optimization: the configuration of the GA and th
e development of an appropriate fitness function, An integer represent
ation for the GA is employed to overcome the difficulty in optimizing
variables that are dissimilar, and the optimal GA configuration is fou
nd through experimental design methods, Three fitness function calcula
tions are compared for their ability to lead the GA to better calibrat
ion models, A fitness function based on the combination of the mean-sq
uared error in the calibration set data, the mean-squared error in the
monitoring set data, and the number of PLS factors raised to a weight
ing factor is found to perform best, Multiple random drawings of the c
alibration and monitoring sets are also found to improve the optimizat
ion performance, Using this fitness function and three random drawings
of the calibration and monitoring sets, the GA found calibration mode
ls that required fewer PLS factors yet had similar or better predictio
n abilities compared to calibration models found through an optimizati
on protocol based on a grid search method.