SPECTRAL BAND SELECTION SUPPORTED BY THE PSO METAHEURISTIC FOR PREDICTION OF ALUMINUM CONTENT OF SOIL SAMPLES

Resumo

Near-infrared (NIR) diffuse reflectance spectroscopy (DRS) is a technique that can be
employed to make a model of soil nutrient prediction by correlating sample data to the respective
reference value obtained by chemical analysis. Such data are organized as attributes of a set of
high-dimensional records and making a model from these data involves difficulties that impair its
performance. To overcome such difficulties, the use of evolutionary algorithms for Feature Selection has
shown to be promising. The aim of this article is to identify, with Particle Swarm Optimization meta
heuristic, the relevant wavelengths for predicting the aluminum content of soil samples from the Campos
Gerais, PR, Brazil region. Results suggest that, for this scenario, few iterations and small swarm size
provide the best subsets.