Under the current international financial situation, how to effectively prevent and control credit risk is a very challenging topic. This paper intends to use the latest machine learning method Extreme Gradient Boosting (XG Boost) to evaluate financial credit. XGBoost method has attracted extensive attention in academic circles because of its efficient and accurate prediction ability. This paper intends to preprocess, screen and optimize the credit information by collecting and sorting it out. On this basis, the credit risk evaluation model is constructed by XGBoost method, and the credit risk is predicted by learning and testing the model. This paper will carry out empirical research in laboratories and real financial institutions to evaluate the effectiveness and robustness of the proposed method. Through comparative analysis, it shows that XGBoost method is superior to traditional methods in accuracy and speed. From the perspective of loan amount, the amount of housing loans is usually higher, such as the ¥ 2 million recorded in the second item and the ¥ 2.5 million recorded in the fifth item, reflecting the large amount of funds required for purchasing a house. The research results of this paper will provide new ideas and methods for risk control in the financial industry. Using XGBoost method to evaluate financial credit risk will help financial institutions make more accurate and scientific decisions in the complex and changing market environment.