CRISP-DM no desenvolvimento de funções de pedotransferência: um estudo de caso com o Banco de Dados HYBRAS

Abstract

This research describes a detailed analysis on the development of pedotransfer functions to estimate Field Capacity and Permanent Wilting Point. The adopted approach utilizes data from the HYBRAS database, which holds information on soil hydraulic constants, and follows the CRISP-DM methodology, providing a standardized framework for model development. The study involves the construction of twelve artificial intelligence models, exploring algorithms that seek both linear and non-linear relationships. The Gradient Boosting algorithm demonstrated the best performance in estimating the Permanent Wilting Point, achieving a coefficient of determination (R²) of 0.74 and a Root Mean Square Error (RMSE) of 0.04 cm³/cm³. The project emphasizes the intention to empower experts during the development of the functions, highlighting the relevance of active participation of these professionals throughout all stages of the process.