Estimativa de atributos químicos dos solos do Cerrado empregando redes neurais artificiais e o algoritmo floresta aleatória

Abstract

Artificial Intelligence (AI) has played a significant role in globalization, profoundly impacting several sectors, including agriculture. In the agricultural context, AI, through techniques such as Artificial Neural Networks (ANNs) and Machine Learning, such as the Random Forest algorithm, has played a crucial role in optimizing agricultural processes, such as crop management, reducing costs and increased precision of operations. In this work, the application of Artificial Neural Networks (ANNs) and the Random Forest (RF) algorithm aimed to estimate soil chemical attributes, aiming to reduce the number of soil samples and compare the methodologies used in training to determine which one obtained the best results in the predictions. With this, a database was selected that contained soil samples with information on calcium (CA), pH, phosphorus (P), potassium (K), magnesium (MG), sum of calcium and magnesium (CA+MG), matter organic matter (MO) and cation exchange capacity (CEC). The study employed three distinct approaches: Model A, Model B and Model C. Models A and B used the Random Forest (RF) algorithm, while Model C employed Artificial Neural Networks (ANNs). Each of these models was trained and validated based on samples from the selected database. The comparative analysis of the models revealed that Model B presented promising results, evaluated using regression model performance metrics, such as the coefficient of determination (R²) and the mean squared error (EMQ). Therefore, it was possible to demonstrate that the application of AI, through techniques such as Artificial Neural Networks (ANNs) and the Random Forest algorithm, had a positive impact on the estimation of soil chemical attributes. Model B stood out as the most effective approach, delivering solid results based on performance assessment metrics.