Predição de microbioma saudável presente na cultura de arroz utilizando técnicas aprendizado de máquina supervisionado

Abstract

Living organisms, including plants, host a wide variety of microorganisms that can have beneficial or detrimental effects on their development. The identification and classification of sets of microorganisms that favor or harm crops are crucial for agricultural development. This study delved into the influence of microorganisms on rice plantations, essential for crop development. Employing neural networks and artificial intelligence algorithms, we classified the health of 110 rice samples, comprising 63 diseased and 47 healthy ones. We utilized the innovative Micronet, the convolutional network MDeep, Decision Tree, Random Forest, MLP, and SVM as classifiers. The results revealed a remarkable ability to identify sets of microorganisms associated with health and disease, enhanced by the use of SHAP values. The ROC curve highlighted the superior performance of Micronet, with an impressive AUC of 94%. MDeep followed closely with 91%, while Random Forest and SVM achieved 88% AUC. The Decision Tree demonstrated solid performance, recording an 83% AUC. These findings suggest that Micronet, especially with the support of SHAP values, is a robust classifier for discerning the health of rice plantations, presenting promising results in the agricultural context. This diversified approach with multiple classifiers highlighted several microorganisms associated with plant health.