To solve the problem of the credit risk control of the small and medium-sized enterprises(SMEs) financing in the supply chain finance field, an ensemble learning model BO-XGBoost-Bagging (BXB) that combines Bayesian optimization and XGBoost was proposed under the Bagging framework. Firstly, based on the XGBoost feature importance, the feature screening was carried out, and the financial credit evaluation index system for the supply chain was established. Secondly, the optimal super parameters of XGBoost were obtained by Bayesian optimization, and the integrated model BXB was obtained by bagging. Finally, the prediction was performed on SMEs data set, and the effectiveness of the credit evaluation model was verified by empirical research. The empirical results show that the BXB model has a better predictive effect than other models and can evaluate the credit risk of SMEs more exactly and comprehensively. The model can better distinguish between risky companies and normal companies, and minimize default losses. It has high application value in the credit evaluation of supply chain finance.