Machine learning algorithms to predict the crops most susceptible to weed occurrence in integrated crop-livestock systems

Abstract

The objective of this work was to investigate the use of machine learning algorithms to predict the crops most susceptible to weed occurrence in integrated crop-livestock systems, based on environmental factors of climate, soil, and cropping systems, to establish correlations between these elements and the occurrence of weeds. Three datasets were used for this purpose: the first provided quantitative information on the invasive species, the second contained data about the soil, and the last had records of the region’s climate. The algorithms used were Support Vector Machine, Decision Tree, Random Forest, and K-Nearest Neighbors. The application of machine learning algorithms to predict the susceptibility of crops to weed emergence is technically feasible and effective. The Decision Tree and Random Forest algorithms demonstrated the best performance, with both models achieving 99% accuracy. Robust relationships were established between environmental factors (climate, soil, and planting) and the appearance of invasive species in certain crops. The algorithms reproduced the patterns of weed emergence observed under field conditions.