Objective To explore the association between pesticide and veterinary drug exposure and the incidence of diabetes mellitus based on Lasso feature selection and the XGBoost model.Methods A cross-sectional study was conducted in Shijiazhuang and Hangzhou， China， between 2018 and 2019， enrolling 80 participants with diabetes and 410 healthy controls. The basic personal information， physical measurements， blood biochemical data， and serum exposure concentration data concerning agricultural or veterinary drugs and chemical pollutants were extracted using a questionnaire. Lasso was used for screening variables. Logical regression and XGBoost models were introduced for data fitting. The area under the curve （AUC） was compared to evaluate the fitting effect， and the variables were ranked by importance.Results Three features from pesticide and veterinary drug exposure， namely PFOA， PFUdA and EHDPP， were selected as related to diabetes， ranking PFOA> EHDPP> PFUdA. The XGBoost model （AUC = 0.83） performed significantly better than the logistic regression model （AUC = 0.64） in this dataset （P<0.05）.Conclusion Lasso is suitable for screening the factors influencing diabetes. The XGBoost model still had a strong ability to fit complex relationships between various influencing factors. PFOA， PFUdA and EHDPP are significant risk factors for diabetes.