Machine-Learning Prediction of Oxidative Stress and Hormonal-Immune Effects from Agrochemical Mixtures in U.S. Farmers

Agrochemicals, such as pesticides, herbicides, and fungicides, are heavily used by U.S. agricultural workers. As a result, they face chronic occupational health risks. There is therefore a need for risk assessment methods that can identify cumulative effects. Furthermore, while machine learning (ML) methods and modelling strategies have great potential to assess these factors, this potential has not yet been analyzed for the biological impacts of farming. The aim of this review is to assess the existing literature on ML and its ability to evaluate oxidative stress, hormonal alterations, and immune system impacts related to occupational exposure to chemical cocktails, specifically as it pertains to farmers in the U.S. Literature was collected from PubMed, Scopus, Web of Science, and Google Scholar from the years 2010 to 2025. To be included, studies had to be peer-reviewed, published within the range of the search, and have used related ML algorithms like mixing models, random forests, ensembles, and neural networks. A total of 8 studies met the inclusion criteria and were included in the final synthesis. In the included studies, machine learning models consistently demonstrated strong predictive performance for oxidative stress biomarkers, hormonal alterations, and immune-related outcomes associated with agrochemical mixture exposure. For toxicity predictions, ML models achieved R² values of 0.75 to 0.89. For predicting interactions, the models achieved an AUC of 0.88 to 0.95. The models also obtained classification accuracies of >80 to 85% for these outcomes. The ML models were able to adequately deal with the complexity of conventional methods in these cases. These values represent performance ranges observed across individual studies rather than pooled quantitative estimates. ML is able to predict effects from oxidative stress, hormonal, and immune effects from agrochemical mixtures, and classifies important synergies and nonlinear effects for agricultural workers. A greater focus on longitudinal data, as well as the use of standard ML protocols, real time data from integrated biomarkers, and hybrid mechanistic and ML models, should allow for personalized data and protective data to provide safer pesticide usage and provide better protective measures for the agricultural workforce in the United States.