Abstract
Vibrational spectroscopy can be used for rapid nutrient assessment of horticultural produce as a means of quality control. Most commonly, spectral data are calibrated against chemical reference data, which are acquired through resource-intensive analytical methods, using partial least squares regression (PLSR). Recently, genetic algorithms (GAs) have been applied to assist PLSR to construct high-performing models through feature selection and latent variable selection. The current approach relies on manually pre-processed data, which requires human expertise and produces inherent biases. To address this limitation, this paper aims to develop a new GA method for automatically selecting the most appropriate pre-processing techniques for specific tasks to bypass manual pre-processing. The results for infrared spectroscopy show the potential of this approach in out-performing manual pre-processing, while the Raman spectroscopy results are competitive, which demonstrates the utility of the approach in terms of saving time and resources.