AUTHOR=Cuenca-Romero Carmen , Apolo-Apolo Orly Enrique , Rodríguez Vázquez Jaime Nolasco , Egea Gregorio , Pérez-Ruiz Manuel 

TITLE=Tackling unbalanced datasets for yellow and brown rust detection in wheat

JOURNAL=Frontiers in Plant Science

VOLUME=15

YEAR=2024

URL=https://www.frontiersin.org/journals/plant-science/articles/10.3389/fpls.2024.1392409

DOI=10.3389/fpls.2024.1392409

ISSN=1664-462X

ABSTRACT=<p>This study evaluates the efficacy of hyperspectral data for detecting yellow and brown rust in wheat, employing machine learning models and the SMOTE (Synthetic Minority Oversampling Technique) augmentation technique to tackle unbalanced datasets. Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest (RF), and Gaussian Naïve Bayes (GNB) models were assessed. Overall, SVM and RF models showed higher accuracies, particularly when utilizing SMOTE-enhanced datasets. The RF model achieved 70% accuracy in detecting yellow rust without data alteration. Conversely, for brown rust, the SVM model outperformed others, reaching 63% accuracy with SMOTE applied to the training set. This study highlights the potential of spectral data and machine learning (ML) techniques in plant disease detection. It emphasizes the need for further research in data processing methodologies, particularly in exploring the impact of techniques like SMOTE on model performance.</p>