Fig. 6: Global mapping of the risk of antibiotic and metal(loid) co-resistant bacteria (AMRB) in agricultural lands.

a Process of building machine learning models. b Evaluation metrics (accuracy, precision, recall, and F1 score) for the best model of each machine learning algorithm on the test set. c Receiver operating characteristic (ROC) curve reflected the classification performance of the best model of random forest model on the test set at different risk levels. d Risk of the antibiotic and metal(loid) co-resistant bacteria (AMRB) in agricultural soils according to machine learning prediction results displayed at a 0.083° resolution using Python. The unsupervised learning approach using k-means clustering was applied to categorize the abundance of AMRB in 511 agricultural soils into six risk levels, which were visualized by the t-distributed stochastic neighbor embedding (t-SNE) method. According to the t-SNE results, a significant gap separates the samples into two groups, in which risk levels of 1, 2, and 3 indicate the low-risk group, and the risk levels of 4, 5, and 6 indicate the high-risk group. Source data were provided as a Source Data file. RF random forest, SVM support vector machine, ANN artificial neural network, KNN K-nearest neighbors, XGBoost: eXtreme Gradient Boosting, ROC receiver operating characteristic.