Predicting Relative Risk of Antimicrobial Resistance using Machine Learning Methods

The extensive use of antibiotics has led to the occurrence of antimicrobial resistance (AMR), which poses a huge threat to human health. There are great challenges in quantifying the absolute risk of AMR since dose-response models for antibiotic resistant bacteria (ARB) and antibiotic resistant genes (ARGs) have not been established. As an alternative, the relative risk analysis of AMR has been already proposed based on global databases. Due to the complicated and cumbersome extraction operations of ARB and ARGs, the relative risk analysis of AMR is still time-consuming and tedious. In this paper, we designed a novel solution of relative risk estimation of AMR based on intelligent methods. For this solution, machine learning methods including logistic regression, decision tree and random forest were applied to rapidly predict the relative risk of AMR in natural aquatic environments when land-use patterns and environmental factors are available as feature variables. We compared the three methods for the AMR relative risk prediction using field-scale drinking-water source data. We found that the random forest method provides approximately 20% improvement in the relative risk prediction of AMR, with accuracy surpassing an 85% threshold. In addition, the importance of features was identified to guide practices. Copyright (C) 2022 The Authors.