Development of genetic algorithm-based wavelength regional selection technique

Abstract

We propose the GAWLS (Genetic Algorithm-based WaveLength Selection) method as a new variable selection technique for spectrum data. It is technique to choose wavelength with important information by a region unit. In this study, we analyzed a NIR spectrum data in precision agriculture with the GAWLS method and confirmed its utility. Q(2) of the GAWLS model was 0.732 and we discovered 3 regions related to quantity of water in the soil. It is expected that we will be able to make stable model with high predictive accuracy by using the GAWLS method, and discovery of informative regions and chemical attribution of constructed model becomes easy.