Application of a Preproceessing Pipeline to VIS-NIR Data for Predicting Soil Nutrient Concentration Values

Eduardo Menezes de Souza Amarante; Julian Santana Liang; Carlos Alberto Campos da Purificação; Rômulo Alexandrino Silva

doi:10.34178/jbth.v7i4.429

Eduardo Menezes de Souza Amarante
Julian Santana Liang
Carlos Alberto Campos da Purificação
Rômulo Alexandrino Silva

DOI: https://doi.org/10.34178/jbth.v7i4.429

Keywords: VIS-NIR Data, Preprocessing Pipeline, Soil Nutrient, Partial Least Square Regression

Abstract

This paper compares the impact of each preprocessing step in predicting soil nutrient concentration values using the partial least squares technique (PLS). The preprocessing pipeline comprises log transformation of the output variable, determination of the optimal number of components, and feature engineering. An increase in the coefficient of determination (R²) and an improvement in model stability were observed.