Abstract: The widespread use of antibiotics in the breeding process has led to the presence of pollutants such as antibiotic resistance genes（ARGs）in livestock manure. At present, the exploration of changes in the abundance of ARGs and mobile genetic elements（MGEs）during aerobic composting of pig manure and their interactions with various factors mainly relies on experimental research, but there is relatively little research on the application of machine learning models. This article collected 191 and 181 valid data on ARGs and MGEs from published articles, respectively, and used support vector machine models to predict the impact of process parameters on ARGs and MGEs. The training set R² of ARGs is 0.986, and the testing set R² is 0.942; The training set R² of MGEs is 0.987, and the testing set R² is 0.921. Using SHAP to analyze target features, it was found that the important features predicted by the model ARGs during aerobic composting are, in order, total composting time, real-time composting temperature, duration of high temperature period, and pH. Analyzing the importance of features in predicting MGEs, the main features are, in order, real-time composting days, real-time composting temperature, total composting time, and duration of high temperature period. This article provides a scientific basis for predicting ARGs and risk control in the process of pig manure treatment.

Key words: antibiotic resistance, machine learning, livestock manure, model explanation