Genetic algorithm-based wavelength selection for the near-infrared determination of glucose in biological matrixes: initialization strategies and effects of spectral resolution.

An improved genetic algorithm (GA)-based wavelength selection procedure is developed to optimize both the near-infrared wavelengths used and the number of latent variables employed in building partial least-squares (PLS) calibration models. This GA-based wavelength selection algorithm is applied to the determination of glucose in two different biological matrixes. With random selection of a small number of initial wavelengths, a dramatic reduction in the number of wavelengths required for building the PLS calibration models is observed. The fitness function used to guide the GA, the method of recombination used, and the effect of spectral resolution on the wavelength selection are also studied. In the resolution study, the original data with a point spacing of 2 cm-1 are deresolved to 4-, 8-, and 16-cm-1 point spacings by truncating the collected interferograms before applying the Fourier processing step. The use of lower resolution spectra is found to reduce further the number of final wavelengths selected by the GA, and the performance of the optimal calibration models obtained with the original spectra is maintained with the lower resolution spectra of both 4- and 8-cm-1 point spacing. Degradation in performance is observed with the spectra computed with a point spacing of 16 cm-1, however.