Use of genetic algorithms in the wavelength selection of FT-MIR spectra to classify unifloral honeys from Sardinia

Beekeeping is among the oldest activities in Sardinia (Italy). Among others, here are produced four valuable unifloral honeys appreciated worldwide for their quality and organoleptic properties, i.e., asphodel (Asphodelus microcarpus), eucalyptus (Eucalyptus camaldulensis), strawberry tree (Arbutus unedo L.) and thistle (Galactites tomentosa). The main purpose of this contribution was to assess a botanical classification method by analyzing 125 honeys using Fourier-Transform Mid-Infrared (FT-MIR) spectroscopy. Spectra were used to develop a predictive model by means of linear discriminant analysis (LDA), using different spectral pretreatments techniques. Predictors were selected using principal component analysis (PCA) or genetic algorithms (GA) tools. In particular, GA selected 34 wavelengths in the spectral regions from 1726 to 1543 cm− 1, and the application of LDA to this selection provided an accuracy of 93.6% in cross validation and an accuracy of 87.8% in the validation on a test set of honey samples. The results were compared, in terms of pros and cons, with other targeted and non-targeted approaches previously assessed by this research group on the same four unifloral honeys.